wisent 0.7.379__py3-none-any.whl → 0.7.701__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- wisent/__init__.py +1 -1
- wisent/core/activations/__init__.py +22 -6
- wisent/core/activations/activations.py +21 -39
- wisent/core/activations/activations_collector.py +141 -373
- wisent/core/activations/classifier_inference_strategy.py +194 -0
- wisent/core/activations/core/atoms.py +8 -92
- wisent/core/activations/extraction_strategy.py +308 -0
- wisent/core/agent/diagnose/response_diagnostics.py +3 -3
- wisent/core/agent/diagnose.py +3 -3
- wisent/core/autonomous_agent.py +2 -2
- wisent/core/cli/agent/apply_steering.py +23 -27
- wisent/core/cli/agent/evaluate_response.py +18 -20
- wisent/core/cli/agent/train_classifier.py +18 -20
- wisent/core/cli/cluster_benchmarks.py +472 -0
- wisent/core/cli/create_steering_vector.py +13 -5
- wisent/core/cli/generate_vector_from_task.py +4 -0
- wisent/core/cli/get_activations.py +12 -36
- wisent/core/cli/method_optimizer.py +859 -0
- wisent/core/cli/optimize.py +44 -5
- wisent/core/cli/optimize_classification.py +5 -6
- wisent/core/cli/optimize_sample_size.py +8 -22
- wisent/core/cli/optimize_steering.py +429 -153
- wisent/core/cli/optimize_weights.py +65 -6
- wisent/core/cli/steering_method_trainer.py +5 -4
- wisent/core/cli/steering_search_space.py +20 -15
- wisent/core/cli/tasks.py +14 -43
- wisent/core/cli/train_unified_goodness.py +17 -18
- wisent/core/contrastive_pairs/diagnostics/control_vectors.py +1578 -173
- wisent/core/contrastive_pairs/diagnostics/linearity.py +63 -80
- wisent/core/contrastive_pairs/diagnostics/vector_quality.py +6 -5
- wisent/core/contrastive_pairs/huggingface_pairs/hf_extractor_manifest.py +5 -19
- wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/__init__.py +11 -5
- wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/apps.py +146 -32
- wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/codexglue.py +2 -2
- wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/humaneval.py +98 -57
- wisent/core/contrastive_pairs/lm_eval_pairs/group_task_manifests/code_x_glue.py +8 -8
- wisent/core/contrastive_pairs/lm_eval_pairs/group_task_manifests/freebase.py +1 -1
- wisent/core/contrastive_pairs/lm_eval_pairs/lm_extractor_manifest.py +8 -5
- wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/agieval_aqua_rat.py +129 -0
- wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/code_x_glue.py +11 -6
- wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/gsm8k.py +1 -1
- wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/mbpp.py +47 -6
- wisent/core/evaluators/benchmark_specific/apps_evaluator.py +133 -0
- wisent/core/evaluators/benchmark_specific/coding/metrics/evaluator.py +6 -1
- wisent/core/evaluators/benchmark_specific/conala_evaluator.py +31 -168
- wisent/core/evaluators/custom/examples/humanization_coherent.py +89 -35
- wisent/core/evaluators/oracles/truthfulqa_gen_evaluator.py +2 -20
- wisent/core/evaluators/personalization/coherence.py +46 -0
- wisent/core/hyperparameter_optimizer.py +13 -13
- wisent/core/lm_eval_harness_ground_truth.py +7 -11
- wisent/core/main.py +3 -0
- wisent/core/models/wisent_model.py +8 -7
- wisent/core/opti/methods/opti_weights.py +29 -2
- wisent/core/optuna/classifier/activation_generator.py +14 -12
- wisent/core/optuna/steering/steering_optimization.py +14 -9
- wisent/core/parser_arguments/cluster_benchmarks_parser.py +31 -0
- wisent/core/parser_arguments/generate_vector_from_task_parser.py +20 -0
- wisent/core/parser_arguments/main_parser.py +8 -0
- wisent/core/parser_arguments/optimize_steering_parser.py +117 -10
- wisent/core/parser_arguments/optimize_weights_parser.py +6 -0
- wisent/core/parser_arguments/tasks_parser.py +7 -19
- wisent/core/steering_methods/core/atoms.py +1 -2
- wisent/core/steering_methods/methods/caa.py +1 -1
- wisent/core/steering_methods/methods/hyperplane.py +74 -0
- wisent/core/steering_methods/methods/prism.py +1 -2
- wisent/core/steering_methods/methods/pulse.py +39 -8
- wisent/core/steering_methods/methods/titan.py +59 -14
- wisent/core/steering_methods/registry.py +52 -12
- wisent/core/steering_optimizer.py +15 -15
- wisent/core/trainers/steering_trainer.py +9 -18
- wisent/parameters/lm_eval/track_progress_not_lm_eval_tasks.json +19 -70
- wisent/scripts/run_quality_metrics_sweep.sh +22 -27
- wisent/tests/test_aggregation_geometry.py +236 -0
- wisent/tests/test_detector_accuracy.py +163 -0
- wisent/tests/test_geometry_exhaustive.py +1202 -0
- wisent/tests/visualize_geometry.py +255 -61
- {wisent-0.7.379.dist-info → wisent-0.7.701.dist-info}/METADATA +1 -1
- {wisent-0.7.379.dist-info → wisent-0.7.701.dist-info}/RECORD +82 -714
- wisent/core/activations/prompt_construction_strategy.py +0 -47
- wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/codexglue_code_to_text.py +0 -15
- wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/codexglue_code_to_text_go.py +0 -64
- wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/codexglue_code_to_text_java.py +0 -65
- wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/codexglue_code_to_text_javascript.py +0 -65
- wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/codexglue_code_to_text_php.py +0 -65
- wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/codexglue_code_to_text_python.py +0 -65
- wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/codexglue_code_to_text_ruby.py +0 -65
- wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/freebase.py +0 -99
- wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/instruct_humaneval.py +0 -180
- wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/instructhumaneval.py +0 -129
- wisent/core/contrastive_pairs/huggingface_pairs/hf_task_extractors/mbpp.py +0 -142
- wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/agieval.py +0 -155
- wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/code2text.py +0 -161
- wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/codexglue.py +0 -107
- wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/livemathbench.py +0 -155
- wisent/core/contrastive_pairs/lm_eval_pairs/lm_task_extractors/polymath.py +0 -155
- wisent/examples/scripts/results/benchmark_descriptions.json +0 -1244
- wisent/examples/scripts/results/benchmark_evaluation_methods.json +0 -66
- wisent/examples/scripts/results/benchmark_evaluator_mapping.json +0 -2781
- wisent/examples/scripts/results/benchmark_evaluator_mapping_updated.json +0 -30536
- wisent/examples/scripts/results/benchmark_evaluators_clean.json +0 -469
- wisent/examples/scripts/results/benchmark_methods_summary.json +0 -260
- wisent/examples/scripts/results/benchmark_pair_creation_methods.json +0 -66
- wisent/examples/scripts/results/benchmark_pair_totals.json +0 -269
- wisent/examples/scripts/results/benchmark_tags.json +0 -917
- wisent/examples/scripts/results/benchmark_test_summary_nov4.json +0 -71
- wisent/examples/scripts/results/coding_benchmarks_test_code_status.json +0 -150
- wisent/examples/scripts/results/failing_benchmarks.json +0 -946
- wisent/examples/scripts/results/failing_benchmarks_list.json +0 -41
- wisent/examples/scripts/results/failing_benchmarks_test_results.json +0 -945
- wisent/examples/scripts/results/missing_benchmark_tags.json +0 -341
- wisent/examples/scripts/results/test_20_newsgroups_evaluation.json +0 -30
- wisent/examples/scripts/results/test_20_newsgroups_pairs.json +0 -8
- wisent/examples/scripts/results/test_AraDICE_evaluation.json +0 -51
- wisent/examples/scripts/results/test_AraDICE_pairs.json +0 -14
- wisent/examples/scripts/results/test_AraDiCE_boolq_egy/test_AraDiCE_boolq_egy_evaluation.json +0 -30
- wisent/examples/scripts/results/test_AraDiCE_boolq_egy/test_AraDiCE_boolq_egy_pairs.json +0 -8
- wisent/examples/scripts/results/test_ArabCulture_evaluation.json +0 -51
- wisent/examples/scripts/results/test_ArabCulture_pairs.json +0 -14
- wisent/examples/scripts/results/test_Tag_evaluation.json +0 -30
- wisent/examples/scripts/results/test_Tag_pairs.json +0 -8
- wisent/examples/scripts/results/test_aclue_evaluation.json +0 -51
- wisent/examples/scripts/results/test_aclue_pairs.json +0 -14
- wisent/examples/scripts/results/test_acp_bench_evaluation.json +0 -51
- wisent/examples/scripts/results/test_acp_bench_hard_evaluation.json +0 -51
- wisent/examples/scripts/results/test_acp_bench_hard_pairs.json +0 -14
- wisent/examples/scripts/results/test_acp_bench_pairs.json +0 -14
- wisent/examples/scripts/results/test_advanced_ai_risk_evaluation.json +0 -51
- wisent/examples/scripts/results/test_advanced_ai_risk_pairs.json +0 -14
- wisent/examples/scripts/results/test_aexams_evaluation.json +0 -51
- wisent/examples/scripts/results/test_aexams_pairs.json +0 -14
- wisent/examples/scripts/results/test_afrimgsm_direct_amh_evaluation.json +0 -30
- wisent/examples/scripts/results/test_afrimgsm_direct_amh_pairs.json +0 -8
- wisent/examples/scripts/results/test_afrimmlu_direct_amh_evaluation.json +0 -30
- wisent/examples/scripts/results/test_afrimmlu_direct_amh_pairs.json +0 -8
- wisent/examples/scripts/results/test_afrixnli_en_direct_amh_evaluation.json +0 -30
- wisent/examples/scripts/results/test_afrixnli_en_direct_amh_pairs.json +0 -8
- wisent/examples/scripts/results/test_ag_news_evaluation.json +0 -30
- wisent/examples/scripts/results/test_ag_news_pairs.json +0 -8
- wisent/examples/scripts/results/test_agieval_evaluation.json +0 -51
- wisent/examples/scripts/results/test_agieval_pairs.json +0 -14
- wisent/examples/scripts/results/test_aime2024_evaluation.json +0 -30
- wisent/examples/scripts/results/test_aime2024_pairs.json +0 -8
- wisent/examples/scripts/results/test_aime2025_evaluation.json +0 -30
- wisent/examples/scripts/results/test_aime2025_pairs.json +0 -8
- wisent/examples/scripts/results/test_aime_evaluation.json +0 -30
- wisent/examples/scripts/results/test_aime_pairs.json +0 -8
- wisent/examples/scripts/results/test_anagrams1_evaluation.json +0 -30
- wisent/examples/scripts/results/test_anagrams1_pairs.json +0 -8
- wisent/examples/scripts/results/test_anagrams2_evaluation.json +0 -30
- wisent/examples/scripts/results/test_anagrams2_pairs.json +0 -8
- wisent/examples/scripts/results/test_anli_evaluation.json +0 -30
- wisent/examples/scripts/results/test_anli_pairs.json +0 -8
- wisent/examples/scripts/results/test_apps_evaluation.json +0 -30
- wisent/examples/scripts/results/test_apps_pairs.json +0 -8
- wisent/examples/scripts/results/test_arabic_exams_evaluation.json +0 -30
- wisent/examples/scripts/results/test_arabic_exams_pairs.json +0 -8
- wisent/examples/scripts/results/test_arabic_leaderboard_complete_evaluation.json +0 -51
- wisent/examples/scripts/results/test_arabic_leaderboard_complete_pairs.json +0 -14
- wisent/examples/scripts/results/test_arabic_leaderboard_light_evaluation.json +0 -51
- wisent/examples/scripts/results/test_arabic_leaderboard_light_pairs.json +0 -14
- wisent/examples/scripts/results/test_arabicmmlu_evaluation.json +0 -51
- wisent/examples/scripts/results/test_arabicmmlu_pairs.json +0 -14
- wisent/examples/scripts/results/test_aradice/test_aradice_evaluation.json +0 -51
- wisent/examples/scripts/results/test_aradice/test_aradice_pairs.json +0 -14
- wisent/examples/scripts/results/test_aradice3/test_aradice_evaluation.json +0 -51
- wisent/examples/scripts/results/test_aradice3/test_aradice_pairs.json +0 -14
- wisent/examples/scripts/results/test_arc_ar_evaluation.json +0 -30
- wisent/examples/scripts/results/test_arc_ar_pairs.json +0 -8
- wisent/examples/scripts/results/test_arc_challenge_evaluation.json +0 -30
- wisent/examples/scripts/results/test_arc_challenge_pairs.json +0 -8
- wisent/examples/scripts/results/test_arc_easy_evaluation.json +0 -30
- wisent/examples/scripts/results/test_arc_easy_pairs.json +0 -8
- wisent/examples/scripts/results/test_argument_topic_evaluation.json +0 -30
- wisent/examples/scripts/results/test_argument_topic_pairs.json +0 -8
- wisent/examples/scripts/results/test_arithmetic_evaluation.json +0 -51
- wisent/examples/scripts/results/test_arithmetic_pairs.json +0 -14
- wisent/examples/scripts/results/test_asdiv_evaluation.json +0 -30
- wisent/examples/scripts/results/test_asdiv_pairs.json +0 -8
- wisent/examples/scripts/results/test_assin_entailment_evaluation.json +0 -30
- wisent/examples/scripts/results/test_assin_entailment_pairs.json +0 -8
- wisent/examples/scripts/results/test_atis_evaluation.json +0 -30
- wisent/examples/scripts/results/test_atis_pairs.json +0 -8
- wisent/examples/scripts/results/test_babi_evaluation.json +0 -30
- wisent/examples/scripts/results/test_babi_pairs.json +0 -8
- wisent/examples/scripts/results/test_babilong_evaluation.json +0 -30
- wisent/examples/scripts/results/test_babilong_pairs.json +0 -8
- wisent/examples/scripts/results/test_bangla_mmlu_evaluation.json +0 -30
- wisent/examples/scripts/results/test_bangla_mmlu_pairs.json +0 -8
- wisent/examples/scripts/results/test_banking77_evaluation.json +0 -30
- wisent/examples/scripts/results/test_banking77_pairs.json +0 -8
- wisent/examples/scripts/results/test_basque/test_basque-glue_pairs.json +0 -14
- wisent/examples/scripts/results/test_basque-glue_evaluation.json +0 -51
- wisent/examples/scripts/results/test_basque-glue_pairs.json +0 -14
- wisent/examples/scripts/results/test_basque2/test_basque-glue_evaluation.json +0 -51
- wisent/examples/scripts/results/test_basque2/test_basque-glue_pairs.json +0 -14
- wisent/examples/scripts/results/test_basque_bench_evaluation.json +0 -51
- wisent/examples/scripts/results/test_basque_bench_pairs.json +0 -14
- wisent/examples/scripts/results/test_basque_glue/test_basque-glue_evaluation.json +0 -51
- wisent/examples/scripts/results/test_basque_glue/test_basque-glue_pairs.json +0 -14
- wisent/examples/scripts/results/test_basqueglue_evaluation.json +0 -51
- wisent/examples/scripts/results/test_basqueglue_pairs.json +0 -14
- wisent/examples/scripts/results/test_bbh_evaluation.json +0 -51
- wisent/examples/scripts/results/test_bbh_pairs.json +0 -14
- wisent/examples/scripts/results/test_bbq_evaluation.json +0 -30
- wisent/examples/scripts/results/test_bbq_pairs.json +0 -8
- wisent/examples/scripts/results/test_bec2016eu_evaluation.json +0 -51
- wisent/examples/scripts/results/test_bec2016eu_pairs.json +0 -14
- wisent/examples/scripts/results/test_belebele_evaluation.json +0 -51
- wisent/examples/scripts/results/test_belebele_pairs.json +0 -14
- wisent/examples/scripts/results/test_benchmarks_evaluation.json +0 -51
- wisent/examples/scripts/results/test_benchmarks_pairs.json +0 -14
- wisent/examples/scripts/results/test_bertaqa_evaluation.json +0 -51
- wisent/examples/scripts/results/test_bertaqa_pairs.json +0 -14
- wisent/examples/scripts/results/test_bhtc_v2_evaluation.json +0 -30
- wisent/examples/scripts/results/test_bhtc_v2_pairs.json +0 -8
- wisent/examples/scripts/results/test_bigbench_evaluation.json +0 -51
- wisent/examples/scripts/results/test_bigbench_pairs.json +0 -14
- wisent/examples/scripts/results/test_blimp_evaluation.json +0 -51
- wisent/examples/scripts/results/test_blimp_pairs.json +0 -14
- wisent/examples/scripts/results/test_boolq/test_boolq_evaluation.json +0 -30
- wisent/examples/scripts/results/test_boolq/test_boolq_pairs.json +0 -8
- wisent/examples/scripts/results/test_boolq-seq2seq_evaluation.json +0 -30
- wisent/examples/scripts/results/test_boolq-seq2seq_pairs.json +0 -8
- wisent/examples/scripts/results/test_boolq_evaluation.json +0 -30
- wisent/examples/scripts/results/test_boolq_pairs.json +0 -8
- wisent/examples/scripts/results/test_c4_evaluation.json +0 -30
- wisent/examples/scripts/results/test_c4_pairs.json +0 -8
- wisent/examples/scripts/results/test_cabreu_evaluation.json +0 -30
- wisent/examples/scripts/results/test_cabreu_pairs.json +0 -8
- wisent/examples/scripts/results/test_careqa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_careqa_pairs.json +0 -8
- wisent/examples/scripts/results/test_catalan_bench_evaluation.json +0 -51
- wisent/examples/scripts/results/test_catalan_bench_pairs.json +0 -14
- wisent/examples/scripts/results/test_catalanqa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_catalanqa_pairs.json +0 -8
- wisent/examples/scripts/results/test_catcola_evaluation.json +0 -30
- wisent/examples/scripts/results/test_catcola_pairs.json +0 -8
- wisent/examples/scripts/results/test_cb_evaluation.json +0 -30
- wisent/examples/scripts/results/test_cb_pairs.json +0 -8
- wisent/examples/scripts/results/test_ceval/test_ceval_evaluation.json +0 -51
- wisent/examples/scripts/results/test_ceval/test_ceval_pairs.json +0 -14
- wisent/examples/scripts/results/test_ceval_accountant/test_ceval-valid_accountant_evaluation.json +0 -30
- wisent/examples/scripts/results/test_ceval_accountant/test_ceval-valid_accountant_pairs.json +0 -8
- wisent/examples/scripts/results/test_ceval_evaluation.json +0 -51
- wisent/examples/scripts/results/test_ceval_pairs.json +0 -14
- wisent/examples/scripts/results/test_ceval_valid/test_ceval_valid_evaluation.json +0 -51
- wisent/examples/scripts/results/test_ceval_valid/test_ceval_valid_pairs.json +0 -14
- wisent/examples/scripts/results/test_chain_of_thought_evaluation.json +0 -51
- wisent/examples/scripts/results/test_chain_of_thought_pairs.json +0 -14
- wisent/examples/scripts/results/test_chartqa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_chartqa_pairs.json +0 -8
- wisent/examples/scripts/results/test_claim_stance_topic_evaluation.json +0 -30
- wisent/examples/scripts/results/test_claim_stance_topic_pairs.json +0 -8
- wisent/examples/scripts/results/test_cmmlu_evaluation.json +0 -51
- wisent/examples/scripts/results/test_cmmlu_pairs.json +0 -14
- wisent/examples/scripts/results/test_cnn_dailymail_evaluation.json +0 -30
- wisent/examples/scripts/results/test_cnn_dailymail_pairs.json +0 -8
- wisent/examples/scripts/results/test_cocoteros_es_evaluation.json +0 -30
- wisent/examples/scripts/results/test_cocoteros_es_pairs.json +0 -8
- wisent/examples/scripts/results/test_codexglue_code_to_text_go_evaluation.json +0 -30
- wisent/examples/scripts/results/test_codexglue_code_to_text_go_pairs.json +0 -8
- wisent/examples/scripts/results/test_codexglue_code_to_text_java_evaluation.json +0 -30
- wisent/examples/scripts/results/test_codexglue_code_to_text_java_pairs.json +0 -8
- wisent/examples/scripts/results/test_codexglue_code_to_text_javascript_evaluation.json +0 -30
- wisent/examples/scripts/results/test_codexglue_code_to_text_javascript_pairs.json +0 -8
- wisent/examples/scripts/results/test_codexglue_code_to_text_php_evaluation.json +0 -30
- wisent/examples/scripts/results/test_codexglue_code_to_text_php_pairs.json +0 -8
- wisent/examples/scripts/results/test_codexglue_code_to_text_python_evaluation.json +0 -30
- wisent/examples/scripts/results/test_codexglue_code_to_text_python_pairs.json +0 -8
- wisent/examples/scripts/results/test_codexglue_code_to_text_ruby_evaluation.json +0 -30
- wisent/examples/scripts/results/test_codexglue_code_to_text_ruby_pairs.json +0 -8
- wisent/examples/scripts/results/test_coedit_gec_evaluation.json +0 -30
- wisent/examples/scripts/results/test_coedit_gec_pairs.json +0 -8
- wisent/examples/scripts/results/test_cola_evaluation.json +0 -30
- wisent/examples/scripts/results/test_cola_pairs.json +0 -8
- wisent/examples/scripts/results/test_commonsense_qa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_commonsense_qa_pairs.json +0 -8
- wisent/examples/scripts/results/test_conala_evaluation.json +0 -30
- wisent/examples/scripts/results/test_conala_pairs.json +0 -8
- wisent/examples/scripts/results/test_concode_evaluation.json +0 -30
- wisent/examples/scripts/results/test_concode_pairs.json +0 -8
- wisent/examples/scripts/results/test_copa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_copa_pairs.json +0 -8
- wisent/examples/scripts/results/test_copal_id_evaluation.json +0 -30
- wisent/examples/scripts/results/test_copal_id_pairs.json +0 -8
- wisent/examples/scripts/results/test_coqa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_coqa_pairs.json +0 -8
- wisent/examples/scripts/results/test_coqcat_evaluation.json +0 -30
- wisent/examples/scripts/results/test_coqcat_pairs.json +0 -8
- wisent/examples/scripts/results/test_crows_pairs_evaluation.json +0 -51
- wisent/examples/scripts/results/test_crows_pairs_pairs.json +0 -14
- wisent/examples/scripts/results/test_csatqa_evaluation.json +0 -51
- wisent/examples/scripts/results/test_csatqa_pairs.json +0 -14
- wisent/examples/scripts/results/test_cycle_letters_evaluation.json +0 -30
- wisent/examples/scripts/results/test_cycle_letters_pairs.json +0 -8
- wisent/examples/scripts/results/test_darija_bench/test_darija_bench_evaluation.json +0 -51
- wisent/examples/scripts/results/test_darija_bench/test_darija_bench_pairs.json +0 -14
- wisent/examples/scripts/results/test_darija_bench_evaluation.json +0 -51
- wisent/examples/scripts/results/test_darija_bench_pairs.json +0 -14
- wisent/examples/scripts/results/test_darijahellaswag_evaluation.json +0 -30
- wisent/examples/scripts/results/test_darijahellaswag_pairs.json +0 -8
- wisent/examples/scripts/results/test_darijammlu_evaluation.json +0 -51
- wisent/examples/scripts/results/test_darijammlu_pairs.json +0 -14
- wisent/examples/scripts/results/test_dbpedia_14_evaluation.json +0 -30
- wisent/examples/scripts/results/test_dbpedia_14_pairs.json +0 -8
- wisent/examples/scripts/results/test_drop_evaluation.json +0 -30
- wisent/examples/scripts/results/test_drop_pairs.json +0 -8
- wisent/examples/scripts/results/test_ds1000_evaluation.json +0 -30
- wisent/examples/scripts/results/test_ds1000_pairs.json +0 -8
- wisent/examples/scripts/results/test_egyhellaswag_evaluation.json +0 -30
- wisent/examples/scripts/results/test_egyhellaswag_pairs.json +0 -8
- wisent/examples/scripts/results/test_egymmlu_evaluation.json +0 -51
- wisent/examples/scripts/results/test_egymmlu_pairs.json +0 -14
- wisent/examples/scripts/results/test_epec_koref_bin_evaluation.json +0 -30
- wisent/examples/scripts/results/test_epec_koref_bin_pairs.json +0 -8
- wisent/examples/scripts/results/test_eq_bench_evaluation.json +0 -30
- wisent/examples/scripts/results/test_eq_bench_pairs.json +0 -8
- wisent/examples/scripts/results/test_escola_evaluation.json +0 -30
- wisent/examples/scripts/results/test_escola_pairs.json +0 -8
- wisent/examples/scripts/results/test_ethics_cm_evaluation.json +0 -30
- wisent/examples/scripts/results/test_ethics_cm_pairs.json +0 -8
- wisent/examples/scripts/results/test_ethos_binary_evaluation.json +0 -30
- wisent/examples/scripts/results/test_ethos_binary_pairs.json +0 -8
- wisent/examples/scripts/results/test_eus_exams/test_eus_exams_evaluation.json +0 -51
- wisent/examples/scripts/results/test_eus_exams/test_eus_exams_pairs.json +0 -14
- wisent/examples/scripts/results/test_eus_exams_es_evaluation.json +0 -51
- wisent/examples/scripts/results/test_eus_exams_es_pairs.json +0 -14
- wisent/examples/scripts/results/test_eus_exams_evaluation.json +0 -51
- wisent/examples/scripts/results/test_eus_exams_pairs.json +0 -14
- wisent/examples/scripts/results/test_eus_proficiency_evaluation.json +0 -30
- wisent/examples/scripts/results/test_eus_proficiency_pairs.json +0 -8
- wisent/examples/scripts/results/test_eus_reading_evaluation.json +0 -30
- wisent/examples/scripts/results/test_eus_reading_pairs.json +0 -8
- wisent/examples/scripts/results/test_eus_trivia_evaluation.json +0 -30
- wisent/examples/scripts/results/test_eus_trivia_pairs.json +0 -8
- wisent/examples/scripts/results/test_evalita-mp_evaluation.json +0 -51
- wisent/examples/scripts/results/test_evalita-mp_pairs.json +0 -14
- wisent/examples/scripts/results/test_evalita-sp_sum_task_fp-small_p1_evaluation.json +0 -30
- wisent/examples/scripts/results/test_evalita-sp_sum_task_fp-small_p1_pairs.json +0 -8
- wisent/examples/scripts/results/test_evalita_LLM_evaluation.json +0 -51
- wisent/examples/scripts/results/test_evalita_LLM_pairs.json +0 -14
- wisent/examples/scripts/results/test_evalita_llm/test_evalita_llm_evaluation.json +0 -51
- wisent/examples/scripts/results/test_evalita_llm/test_evalita_llm_pairs.json +0 -14
- wisent/examples/scripts/results/test_evalita_mp/test_evalita-mp_te_prompt-1_evaluation.json +0 -30
- wisent/examples/scripts/results/test_evalita_mp/test_evalita-mp_te_prompt-1_pairs.json +0 -8
- wisent/examples/scripts/results/test_evalita_mp2/test_evalita_mp_evaluation.json +0 -51
- wisent/examples/scripts/results/test_evalita_mp2/test_evalita_mp_pairs.json +0 -14
- wisent/examples/scripts/results/test_evalita_sp2/test_evalita-sp_sum_task_fp-small_p1_evaluation.json +0 -30
- wisent/examples/scripts/results/test_evalita_sp2/test_evalita-sp_sum_task_fp-small_p1_pairs.json +0 -8
- wisent/examples/scripts/results/test_fda_evaluation.json +0 -30
- wisent/examples/scripts/results/test_fda_pairs.json +0 -8
- wisent/examples/scripts/results/test_financial_tweets_evaluation.json +0 -30
- wisent/examples/scripts/results/test_financial_tweets_pairs.json +0 -8
- wisent/examples/scripts/results/test_fld/test_fld_evaluation.json +0 -30
- wisent/examples/scripts/results/test_fld/test_fld_pairs.json +0 -8
- wisent/examples/scripts/results/test_fld_evaluation.json +0 -30
- wisent/examples/scripts/results/test_fld_fixed/test_fld_evaluation.json +0 -30
- wisent/examples/scripts/results/test_fld_fixed/test_fld_pairs.json +0 -8
- wisent/examples/scripts/results/test_fld_pairs.json +0 -8
- wisent/examples/scripts/results/test_flores_evaluation.json +0 -51
- wisent/examples/scripts/results/test_flores_pairs.json +0 -14
- wisent/examples/scripts/results/test_freebase_evaluation.json +0 -30
- wisent/examples/scripts/results/test_freebase_pairs.json +0 -8
- wisent/examples/scripts/results/test_french_bench_evaluation.json +0 -51
- wisent/examples/scripts/results/test_french_bench_pairs.json +0 -14
- wisent/examples/scripts/results/test_galcola_evaluation.json +0 -30
- wisent/examples/scripts/results/test_galcola_pairs.json +0 -8
- wisent/examples/scripts/results/test_galician_bench_evaluation.json +0 -51
- wisent/examples/scripts/results/test_galician_bench_pairs.json +0 -14
- wisent/examples/scripts/results/test_glianorex_evaluation.json +0 -30
- wisent/examples/scripts/results/test_glianorex_pairs.json +0 -8
- wisent/examples/scripts/results/test_global_mmlu_evaluation.json +0 -51
- wisent/examples/scripts/results/test_global_mmlu_pairs.json +0 -14
- wisent/examples/scripts/results/test_glue_evaluation.json +0 -51
- wisent/examples/scripts/results/test_glue_pairs.json +0 -14
- wisent/examples/scripts/results/test_gpqa_evaluation.json +0 -51
- wisent/examples/scripts/results/test_gpqa_pairs.json +0 -14
- wisent/examples/scripts/results/test_gpt3_translation_benchmarks_evaluation.json +0 -51
- wisent/examples/scripts/results/test_gpt3_translation_benchmarks_pairs.json +0 -14
- wisent/examples/scripts/results/test_groundcocoa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_groundcocoa_pairs.json +0 -8
- wisent/examples/scripts/results/test_gsm8k_evaluation.json +0 -30
- wisent/examples/scripts/results/test_gsm8k_pairs.json +0 -8
- wisent/examples/scripts/results/test_haerae_evaluation.json +0 -51
- wisent/examples/scripts/results/test_haerae_pairs.json +0 -14
- wisent/examples/scripts/results/test_headqa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_headqa_pairs.json +0 -8
- wisent/examples/scripts/results/test_hellaswag_evaluation.json +0 -30
- wisent/examples/scripts/results/test_hellaswag_pairs.json +0 -8
- wisent/examples/scripts/results/test_hendrycks_ethics_evaluation.json +0 -51
- wisent/examples/scripts/results/test_hendrycks_ethics_pairs.json +0 -14
- wisent/examples/scripts/results/test_hendrycks_math_evaluation.json +0 -51
- wisent/examples/scripts/results/test_hendrycks_math_pairs.json +0 -14
- wisent/examples/scripts/results/test_histoires_morales_evaluation.json +0 -30
- wisent/examples/scripts/results/test_histoires_morales_pairs.json +0 -8
- wisent/examples/scripts/results/test_hmmt_evaluation.json +0 -30
- wisent/examples/scripts/results/test_hmmt_feb_2025_evaluation.json +0 -30
- wisent/examples/scripts/results/test_hmmt_feb_2025_pairs.json +0 -8
- wisent/examples/scripts/results/test_hmmt_pairs.json +0 -8
- wisent/examples/scripts/results/test_hrm8k_evaluation.json +0 -51
- wisent/examples/scripts/results/test_hrm8k_pairs.json +0 -14
- wisent/examples/scripts/results/test_humaneval_evaluation.json +0 -30
- wisent/examples/scripts/results/test_humaneval_pairs.json +0 -8
- wisent/examples/scripts/results/test_humaneval_plus_evaluation.json +0 -30
- wisent/examples/scripts/results/test_humaneval_plus_pairs.json +0 -8
- wisent/examples/scripts/results/test_ifeval_evaluation.json +0 -30
- wisent/examples/scripts/results/test_ifeval_pairs.json +0 -8
- wisent/examples/scripts/results/test_instruct_humaneval/test_instruct_humaneval_evaluation.json +0 -30
- wisent/examples/scripts/results/test_instruct_humaneval/test_instruct_humaneval_pairs.json +0 -8
- wisent/examples/scripts/results/test_instruct_humaneval_evaluation.json +0 -30
- wisent/examples/scripts/results/test_instruct_humaneval_pairs.json +0 -8
- wisent/examples/scripts/results/test_inverse_scaling_evaluation.json +0 -51
- wisent/examples/scripts/results/test_inverse_scaling_hindsight_neglect_10shot_evaluation.json +0 -30
- wisent/examples/scripts/results/test_inverse_scaling_hindsight_neglect_10shot_pairs.json +0 -8
- wisent/examples/scripts/results/test_inverse_scaling_mc/test_inverse_scaling_mc_evaluation.json +0 -51
- wisent/examples/scripts/results/test_inverse_scaling_mc/test_inverse_scaling_mc_pairs.json +0 -14
- wisent/examples/scripts/results/test_inverse_scaling_pairs.json +0 -14
- wisent/examples/scripts/results/test_iwslt2017-ar-en_evaluation.json +0 -30
- wisent/examples/scripts/results/test_iwslt2017-ar-en_pairs.json +0 -8
- wisent/examples/scripts/results/test_iwslt2017-en-ar_evaluation.json +0 -30
- wisent/examples/scripts/results/test_iwslt2017-en-ar_pairs.json +0 -8
- wisent/examples/scripts/results/test_iwslt2017_ar_en/test_iwslt2017-ar-en_evaluation.json +0 -30
- wisent/examples/scripts/results/test_iwslt2017_ar_en/test_iwslt2017-ar-en_pairs.json +0 -8
- wisent/examples/scripts/results/test_iwslt2017_en_ar/test_iwslt2017-en-ar_evaluation.json +0 -30
- wisent/examples/scripts/results/test_iwslt2017_en_ar/test_iwslt2017-en-ar_pairs.json +0 -8
- wisent/examples/scripts/results/test_iwslt2017_group/test_iwslt2017_evaluation.json +0 -30
- wisent/examples/scripts/results/test_iwslt2017_group/test_iwslt2017_pairs.json +0 -8
- wisent/examples/scripts/results/test_japanese_leaderboard_evaluation.json +0 -51
- wisent/examples/scripts/results/test_japanese_leaderboard_pairs.json +0 -14
- wisent/examples/scripts/results/test_jsonschema_bench/test_jsonschema_bench_evaluation.json +0 -30
- wisent/examples/scripts/results/test_jsonschema_bench/test_jsonschema_bench_pairs.json +0 -8
- wisent/examples/scripts/results/test_jsonschema_bench_evaluation.json +0 -30
- wisent/examples/scripts/results/test_jsonschema_bench_final/test_jsonschema_bench_evaluation.json +0 -30
- wisent/examples/scripts/results/test_jsonschema_bench_final/test_jsonschema_bench_pairs.json +0 -8
- wisent/examples/scripts/results/test_jsonschema_bench_pairs.json +0 -8
- wisent/examples/scripts/results/test_kbl_evaluation.json +0 -51
- wisent/examples/scripts/results/test_kbl_fixed/test_kbl_evaluation.json +0 -51
- wisent/examples/scripts/results/test_kbl_fixed/test_kbl_pairs.json +0 -14
- wisent/examples/scripts/results/test_kbl_pairs.json +0 -14
- wisent/examples/scripts/results/test_kmmlu_evaluation.json +0 -51
- wisent/examples/scripts/results/test_kmmlu_pairs.json +0 -14
- wisent/examples/scripts/results/test_kobest_evaluation.json +0 -51
- wisent/examples/scripts/results/test_kobest_pairs.json +0 -14
- wisent/examples/scripts/results/test_kormedmcqa/test_kormedmcqa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_kormedmcqa/test_kormedmcqa_pairs.json +0 -8
- wisent/examples/scripts/results/test_kormedmcqa_dentist/test_kormedmcqa_dentist_evaluation.json +0 -30
- wisent/examples/scripts/results/test_kormedmcqa_dentist/test_kormedmcqa_dentist_pairs.json +0 -8
- wisent/examples/scripts/results/test_kormedmcqa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_kormedmcqa_pairs.json +0 -8
- wisent/examples/scripts/results/test_lambada_cloze_evaluation.json +0 -30
- wisent/examples/scripts/results/test_lambada_cloze_pairs.json +0 -8
- wisent/examples/scripts/results/test_lambada_evaluation.json +0 -30
- wisent/examples/scripts/results/test_lambada_final/test_lambada_openai_mt_stablelm_en_evaluation.json +0 -30
- wisent/examples/scripts/results/test_lambada_final/test_lambada_openai_mt_stablelm_en_pairs.json +0 -8
- wisent/examples/scripts/results/test_lambada_multilingual/test_lambada_multilingual_evaluation.json +0 -51
- wisent/examples/scripts/results/test_lambada_multilingual/test_lambada_multilingual_pairs.json +0 -14
- wisent/examples/scripts/results/test_lambada_multilingual_evaluation.json +0 -51
- wisent/examples/scripts/results/test_lambada_multilingual_pairs.json +0 -14
- wisent/examples/scripts/results/test_lambada_multilingual_stablelm_evaluation.json +0 -51
- wisent/examples/scripts/results/test_lambada_multilingual_stablelm_pairs.json +0 -14
- wisent/examples/scripts/results/test_lambada_openai_evaluation.json +0 -30
- wisent/examples/scripts/results/test_lambada_openai_pairs.json +0 -8
- wisent/examples/scripts/results/test_lambada_pairs.json +0 -8
- wisent/examples/scripts/results/test_lambada_stablelm_en_fixed/test_lambada_openai_mt_stablelm_en_evaluation.json +0 -30
- wisent/examples/scripts/results/test_lambada_stablelm_en_fixed/test_lambada_openai_mt_stablelm_en_pairs.json +0 -8
- wisent/examples/scripts/results/test_lambada_stablelm_fixed/test_lambada_openai_mt_stablelm_en_evaluation.json +0 -30
- wisent/examples/scripts/results/test_lambada_stablelm_fixed/test_lambada_openai_mt_stablelm_en_pairs.json +0 -8
- wisent/examples/scripts/results/test_lambada_standard_evaluation.json +0 -30
- wisent/examples/scripts/results/test_lambada_standard_pairs.json +0 -8
- wisent/examples/scripts/results/test_leaderboard_evaluation.json +0 -51
- wisent/examples/scripts/results/test_leaderboard_pairs.json +0 -14
- wisent/examples/scripts/results/test_libra/test_libra_evaluation.json +0 -51
- wisent/examples/scripts/results/test_libra/test_libra_pairs.json +0 -14
- wisent/examples/scripts/results/test_libra_evaluation.json +0 -51
- wisent/examples/scripts/results/test_libra_pairs.json +0 -14
- wisent/examples/scripts/results/test_lingoly_evaluation.json +0 -30
- wisent/examples/scripts/results/test_lingoly_pairs.json +0 -8
- wisent/examples/scripts/results/test_livecodebench_evaluation.json +0 -30
- wisent/examples/scripts/results/test_livecodebench_pairs.json +0 -8
- wisent/examples/scripts/results/test_livemathbench_cnmo_en_evaluation.json +0 -30
- wisent/examples/scripts/results/test_livemathbench_cnmo_en_pairs.json +0 -8
- wisent/examples/scripts/results/test_livemathbench_cnmo_zh_evaluation.json +0 -30
- wisent/examples/scripts/results/test_livemathbench_cnmo_zh_pairs.json +0 -8
- wisent/examples/scripts/results/test_llama_evaluation.json +0 -30
- wisent/examples/scripts/results/test_llama_pairs.json +0 -8
- wisent/examples/scripts/results/test_logiqa2_evaluation.json +0 -30
- wisent/examples/scripts/results/test_logiqa2_pairs.json +0 -8
- wisent/examples/scripts/results/test_logiqa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_logiqa_pairs.json +0 -8
- wisent/examples/scripts/results/test_m_mmlu_evaluation.json +0 -51
- wisent/examples/scripts/results/test_m_mmlu_pairs.json +0 -14
- wisent/examples/scripts/results/test_mastermind/test_mastermind_evaluation.json +0 -51
- wisent/examples/scripts/results/test_mastermind/test_mastermind_pairs.json +0 -14
- wisent/examples/scripts/results/test_mastermind_24_easy/test_mastermind_24_easy_evaluation.json +0 -30
- wisent/examples/scripts/results/test_mastermind_24_easy/test_mastermind_24_easy_pairs.json +0 -8
- wisent/examples/scripts/results/test_mastermind_evaluation.json +0 -51
- wisent/examples/scripts/results/test_mastermind_pairs.json +0 -14
- wisent/examples/scripts/results/test_math500_evaluation.json +0 -30
- wisent/examples/scripts/results/test_math500_pairs.json +0 -8
- wisent/examples/scripts/results/test_math_evaluation.json +0 -30
- wisent/examples/scripts/results/test_math_pairs.json +0 -8
- wisent/examples/scripts/results/test_mathqa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_mathqa_pairs.json +0 -8
- wisent/examples/scripts/results/test_mbpp_evaluation.json +0 -30
- wisent/examples/scripts/results/test_mbpp_pairs.json +0 -8
- wisent/examples/scripts/results/test_mbpp_plus_evaluation.json +0 -30
- wisent/examples/scripts/results/test_mbpp_plus_pairs.json +0 -8
- wisent/examples/scripts/results/test_mc_taco_evaluation.json +0 -30
- wisent/examples/scripts/results/test_mc_taco_pairs.json +0 -8
- wisent/examples/scripts/results/test_med_concepts_qa/test_med_concepts_qa_evaluation.json +0 -51
- wisent/examples/scripts/results/test_med_concepts_qa/test_med_concepts_qa_pairs.json +0 -14
- wisent/examples/scripts/results/test_med_concepts_qa_atc_easy/test_med_concepts_qa_atc_easy_evaluation.json +0 -30
- wisent/examples/scripts/results/test_med_concepts_qa_atc_easy/test_med_concepts_qa_atc_easy_pairs.json +0 -8
- wisent/examples/scripts/results/test_med_concepts_qa_evaluation.json +0 -51
- wisent/examples/scripts/results/test_med_concepts_qa_pairs.json +0 -14
- wisent/examples/scripts/results/test_meddialog_evaluation.json +0 -30
- wisent/examples/scripts/results/test_meddialog_pairs.json +0 -8
- wisent/examples/scripts/results/test_meddialog_raw_perplexity/test_meddialog_raw_perplexity_evaluation.json +0 -30
- wisent/examples/scripts/results/test_meddialog_raw_perplexity/test_meddialog_raw_perplexity_pairs.json +0 -8
- wisent/examples/scripts/results/test_mediqa_qa2019_evaluation.json +0 -30
- wisent/examples/scripts/results/test_mediqa_qa2019_pairs.json +0 -8
- wisent/examples/scripts/results/test_medmcqa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_medmcqa_pairs.json +0 -8
- wisent/examples/scripts/results/test_medqa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_medqa_pairs.json +0 -8
- wisent/examples/scripts/results/test_medtext_evaluation.json +0 -30
- wisent/examples/scripts/results/test_medtext_pairs.json +0 -8
- wisent/examples/scripts/results/test_mela_evaluation.json +0 -51
- wisent/examples/scripts/results/test_mela_pairs.json +0 -14
- wisent/examples/scripts/results/test_meqsum_evaluation.json +0 -30
- wisent/examples/scripts/results/test_meqsum_pairs.json +0 -8
- wisent/examples/scripts/results/test_mercury_evaluation.json +0 -30
- wisent/examples/scripts/results/test_mercury_pairs.json +0 -8
- wisent/examples/scripts/results/test_metabench_evaluation.json +0 -51
- wisent/examples/scripts/results/test_metabench_pairs.json +0 -14
- wisent/examples/scripts/results/test_mgsm_evaluation.json +0 -51
- wisent/examples/scripts/results/test_mgsm_pairs.json +0 -14
- wisent/examples/scripts/results/test_mimic_repsum_evaluation.json +0 -30
- wisent/examples/scripts/results/test_mimic_repsum_pairs.json +0 -8
- wisent/examples/scripts/results/test_minerva_math_evaluation.json +0 -51
- wisent/examples/scripts/results/test_minerva_math_pairs.json +0 -14
- wisent/examples/scripts/results/test_mlqa_evaluation.json +0 -51
- wisent/examples/scripts/results/test_mlqa_pairs.json +0 -14
- wisent/examples/scripts/results/test_mmlu-pro-plus_evaluation.json +0 -51
- wisent/examples/scripts/results/test_mmlu-pro-plus_pairs.json +0 -14
- wisent/examples/scripts/results/test_mmlu_evaluation.json +0 -51
- wisent/examples/scripts/results/test_mmlu_pairs.json +0 -14
- wisent/examples/scripts/results/test_mmlu_pro_evaluation.json +0 -51
- wisent/examples/scripts/results/test_mmlu_pro_pairs.json +0 -14
- wisent/examples/scripts/results/test_mmlu_prox_evaluation.json +0 -51
- wisent/examples/scripts/results/test_mmlu_prox_pairs.json +0 -14
- wisent/examples/scripts/results/test_mmlusr_evaluation.json +0 -30
- wisent/examples/scripts/results/test_mmlusr_pairs.json +0 -8
- wisent/examples/scripts/results/test_mmmu_evaluation.json +0 -51
- wisent/examples/scripts/results/test_mmmu_pairs.json +0 -14
- wisent/examples/scripts/results/test_mnli_evaluation.json +0 -30
- wisent/examples/scripts/results/test_mnli_pairs.json +0 -8
- wisent/examples/scripts/results/test_model_written_evals_evaluation.json +0 -51
- wisent/examples/scripts/results/test_model_written_evals_pairs.json +0 -14
- wisent/examples/scripts/results/test_moral_stories_evaluation.json +0 -30
- wisent/examples/scripts/results/test_moral_stories_pairs.json +0 -8
- wisent/examples/scripts/results/test_mts_dialog_evaluation.json +0 -30
- wisent/examples/scripts/results/test_mts_dialog_pairs.json +0 -8
- wisent/examples/scripts/results/test_multiblimp_evaluation.json +0 -51
- wisent/examples/scripts/results/test_multiblimp_pairs.json +0 -14
- wisent/examples/scripts/results/test_multimedqa_evaluation.json +0 -51
- wisent/examples/scripts/results/test_multimedqa_pairs.json +0 -14
- wisent/examples/scripts/results/test_multipl_e_evaluation.json +0 -30
- wisent/examples/scripts/results/test_multipl_e_pairs.json +0 -8
- wisent/examples/scripts/results/test_mutual_evaluation.json +0 -30
- wisent/examples/scripts/results/test_mutual_pairs.json +0 -8
- wisent/examples/scripts/results/test_non_greedy_robustness_agieval_aqua_rat_evaluation.json +0 -30
- wisent/examples/scripts/results/test_non_greedy_robustness_agieval_aqua_rat_pairs.json +0 -8
- wisent/examples/scripts/results/test_noreval_evaluation.json +0 -51
- wisent/examples/scripts/results/test_noreval_pairs.json +0 -14
- wisent/examples/scripts/results/test_noticia_evaluation.json +0 -30
- wisent/examples/scripts/results/test_noticia_pairs.json +0 -8
- wisent/examples/scripts/results/test_nq_open_evaluation.json +0 -30
- wisent/examples/scripts/results/test_nq_open_pairs.json +0 -8
- wisent/examples/scripts/results/test_olaph_evaluation.json +0 -30
- wisent/examples/scripts/results/test_olaph_pairs.json +0 -8
- wisent/examples/scripts/results/test_openbookqa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_openbookqa_pairs.json +0 -8
- wisent/examples/scripts/results/test_openllm_evaluation.json +0 -51
- wisent/examples/scripts/results/test_openllm_pairs.json +0 -14
- wisent/examples/scripts/results/test_option_order_robustness_agieval_aqua_rat_evaluation.json +0 -30
- wisent/examples/scripts/results/test_option_order_robustness_agieval_aqua_rat_pairs.json +0 -8
- wisent/examples/scripts/results/test_paloma_evaluation.json +0 -51
- wisent/examples/scripts/results/test_paloma_pairs.json +0 -14
- wisent/examples/scripts/results/test_passkey/test_passkey_evaluation.json +0 -30
- wisent/examples/scripts/results/test_passkey/test_passkey_pairs.json +0 -8
- wisent/examples/scripts/results/test_paws-x_evaluation.json +0 -51
- wisent/examples/scripts/results/test_paws-x_pairs.json +0 -14
- wisent/examples/scripts/results/test_paws_en/test_paws_en_evaluation.json +0 -30
- wisent/examples/scripts/results/test_paws_en/test_paws_en_pairs.json +0 -8
- wisent/examples/scripts/results/test_penn_treebank_evaluation.json +0 -30
- wisent/examples/scripts/results/test_penn_treebank_pairs.json +0 -8
- wisent/examples/scripts/results/test_pile_10k/test_pile_10k_evaluation.json +0 -30
- wisent/examples/scripts/results/test_pile_10k/test_pile_10k_pairs.json +0 -8
- wisent/examples/scripts/results/test_piqa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_piqa_pairs.json +0 -8
- wisent/examples/scripts/results/test_polemo2_evaluation.json +0 -30
- wisent/examples/scripts/results/test_polemo2_pairs.json +0 -8
- wisent/examples/scripts/results/test_polymath_en_high_evaluation.json +0 -30
- wisent/examples/scripts/results/test_polymath_en_high_pairs.json +0 -8
- wisent/examples/scripts/results/test_polymath_en_medium_evaluation.json +0 -30
- wisent/examples/scripts/results/test_polymath_en_medium_pairs.json +0 -8
- wisent/examples/scripts/results/test_polymath_zh_high_evaluation.json +0 -30
- wisent/examples/scripts/results/test_polymath_zh_high_pairs.json +0 -8
- wisent/examples/scripts/results/test_polymath_zh_medium_evaluation.json +0 -30
- wisent/examples/scripts/results/test_polymath_zh_medium_pairs.json +0 -8
- wisent/examples/scripts/results/test_portuguese_bench_evaluation.json +0 -51
- wisent/examples/scripts/results/test_portuguese_bench_pairs.json +0 -14
- wisent/examples/scripts/results/test_prompt_robustness_agieval_aqua_rat/test_prompt_robustness_agieval_aqua_rat_evaluation.json +0 -30
- wisent/examples/scripts/results/test_prompt_robustness_agieval_aqua_rat/test_prompt_robustness_agieval_aqua_rat_pairs.json +0 -8
- wisent/examples/scripts/results/test_prompt_robustness_agieval_aqua_rat_evaluation.json +0 -30
- wisent/examples/scripts/results/test_prompt_robustness_agieval_aqua_rat_pairs.json +0 -8
- wisent/examples/scripts/results/test_prost_evaluation.json +0 -30
- wisent/examples/scripts/results/test_prost_pairs.json +0 -8
- wisent/examples/scripts/results/test_ptb_evaluation.json +0 -30
- wisent/examples/scripts/results/test_ptb_pairs.json +0 -8
- wisent/examples/scripts/results/test_pubmedqa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_pubmedqa_pairs.json +0 -8
- wisent/examples/scripts/results/test_pythia_evaluation.json +0 -51
- wisent/examples/scripts/results/test_pythia_pairs.json +0 -14
- wisent/examples/scripts/results/test_qa4mre_evaluation.json +0 -30
- wisent/examples/scripts/results/test_qa4mre_pairs.json +0 -8
- wisent/examples/scripts/results/test_qasper_evaluation.json +0 -30
- wisent/examples/scripts/results/test_qasper_pairs.json +0 -8
- wisent/examples/scripts/results/test_race_evaluation.json +0 -30
- wisent/examples/scripts/results/test_race_pairs.json +0 -8
- wisent/examples/scripts/results/test_realtoxicityprompts_evaluation.json +0 -30
- wisent/examples/scripts/results/test_realtoxicityprompts_pairs.json +0 -8
- wisent/examples/scripts/results/test_recode_evaluation.json +0 -30
- wisent/examples/scripts/results/test_recode_pairs.json +0 -8
- wisent/examples/scripts/results/test_record_evaluation.json +0 -30
- wisent/examples/scripts/results/test_record_pairs.json +0 -8
- wisent/examples/scripts/results/test_ruler_evaluation.json +0 -51
- wisent/examples/scripts/results/test_ruler_pairs.json +0 -14
- wisent/examples/scripts/results/test_sciq_evaluation.json +0 -30
- wisent/examples/scripts/results/test_sciq_pairs.json +0 -8
- wisent/examples/scripts/results/test_score_evaluation.json +0 -51
- wisent/examples/scripts/results/test_score_pairs.json +0 -14
- wisent/examples/scripts/results/test_self_consistency_evaluation.json +0 -30
- wisent/examples/scripts/results/test_self_consistency_pairs.json +0 -8
- wisent/examples/scripts/results/test_siqa/test_siqa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_siqa/test_siqa_pairs.json +0 -8
- wisent/examples/scripts/results/test_siqa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_siqa_pairs.json +0 -8
- wisent/examples/scripts/results/test_spanish_bench_evaluation.json +0 -51
- wisent/examples/scripts/results/test_spanish_bench_pairs.json +0 -14
- wisent/examples/scripts/results/test_squad2_evaluation.json +0 -30
- wisent/examples/scripts/results/test_squad2_pairs.json +0 -8
- wisent/examples/scripts/results/test_squadv2_evaluation.json +0 -30
- wisent/examples/scripts/results/test_squadv2_pairs.json +0 -8
- wisent/examples/scripts/results/test_super-glue-lm-eval-v1-seq2seq_evaluation.json +0 -30
- wisent/examples/scripts/results/test_super-glue-lm-eval-v1-seq2seq_pairs.json +0 -8
- wisent/examples/scripts/results/test_super-glue-lm-eval-v1_evaluation.json +0 -51
- wisent/examples/scripts/results/test_super-glue-lm-eval-v1_pairs.json +0 -14
- wisent/examples/scripts/results/test_swag_evaluation.json +0 -30
- wisent/examples/scripts/results/test_swag_pairs.json +0 -8
- wisent/examples/scripts/results/test_tinyBenchmarks_evaluation.json +0 -51
- wisent/examples/scripts/results/test_tinyBenchmarks_pairs.json +0 -14
- wisent/examples/scripts/results/test_tmmluplus_evaluation.json +0 -51
- wisent/examples/scripts/results/test_tmmluplus_pairs.json +0 -14
- wisent/examples/scripts/results/test_translation_evaluation.json +0 -51
- wisent/examples/scripts/results/test_translation_pairs.json +0 -14
- wisent/examples/scripts/results/test_triviaqa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_triviaqa_pairs.json +0 -8
- wisent/examples/scripts/results/test_truthfulqa-multi_evaluation.json +0 -51
- wisent/examples/scripts/results/test_truthfulqa-multi_pairs.json +0 -14
- wisent/examples/scripts/results/test_truthfulqa_evaluation.json +0 -30
- wisent/examples/scripts/results/test_truthfulqa_mc1_evaluation.json +0 -30
- wisent/examples/scripts/results/test_truthfulqa_mc1_pairs.json +0 -8
- wisent/examples/scripts/results/test_truthfulqa_mc2_evaluation.json +0 -30
- wisent/examples/scripts/results/test_truthfulqa_mc2_pairs.json +0 -8
- wisent/examples/scripts/results/test_truthfulqa_pairs.json +0 -8
- wisent/examples/scripts/results/test_turkishmmlu_evaluation.json +0 -51
- wisent/examples/scripts/results/test_turkishmmlu_pairs.json +0 -14
- wisent/examples/scripts/results/test_unfair_tos_evaluation.json +0 -30
- wisent/examples/scripts/results/test_unfair_tos_pairs.json +0 -8
- wisent/examples/scripts/results/test_unscramble_evaluation.json +0 -51
- wisent/examples/scripts/results/test_unscramble_pairs.json +0 -14
- wisent/examples/scripts/results/test_webqs_evaluation.json +0 -30
- wisent/examples/scripts/results/test_webqs_pairs.json +0 -8
- wisent/examples/scripts/results/test_wikitext103_evaluation.json +0 -30
- wisent/examples/scripts/results/test_wikitext103_pairs.json +0 -8
- wisent/examples/scripts/results/test_wikitext_evaluation.json +0 -30
- wisent/examples/scripts/results/test_wikitext_pairs.json +0 -8
- wisent/examples/scripts/results/test_winogender_evaluation.json +0 -51
- wisent/examples/scripts/results/test_winogender_pairs.json +0 -14
- wisent/examples/scripts/results/test_winogrande_evaluation.json +0 -30
- wisent/examples/scripts/results/test_winogrande_pairs.json +0 -8
- wisent/examples/scripts/results/test_wmdp_evaluation.json +0 -30
- wisent/examples/scripts/results/test_wmdp_pairs.json +0 -8
- wisent/examples/scripts/results/test_wmt-ro-en-t5-prompt_evaluation.json +0 -30
- wisent/examples/scripts/results/test_wmt-ro-en-t5-prompt_pairs.json +0 -8
- wisent/examples/scripts/results/test_wmt14_en_fr_evaluation.json +0 -30
- wisent/examples/scripts/results/test_wmt14_en_fr_pairs.json +0 -8
- wisent/examples/scripts/results/test_wmt16_en_de_evaluation.json +0 -30
- wisent/examples/scripts/results/test_wmt16_en_de_pairs.json +0 -8
- wisent/examples/scripts/results/test_wmt16_ro_en_evaluation.json +0 -30
- wisent/examples/scripts/results/test_wmt16_ro_en_pairs.json +0 -8
- wisent/examples/scripts/results/test_wsc273_evaluation.json +0 -30
- wisent/examples/scripts/results/test_wsc273_pairs.json +0 -8
- wisent/examples/scripts/results/test_xcopa_evaluation.json +0 -51
- wisent/examples/scripts/results/test_xcopa_pairs.json +0 -14
- wisent/examples/scripts/results/test_xnli_eu_evaluation.json +0 -30
- wisent/examples/scripts/results/test_xnli_eu_pairs.json +0 -8
- wisent/examples/scripts/results/test_xnli_evaluation.json +0 -51
- wisent/examples/scripts/results/test_xnli_pairs.json +0 -14
- wisent/examples/scripts/results/test_xquad_evaluation.json +0 -51
- wisent/examples/scripts/results/test_xquad_pairs.json +0 -14
- wisent/examples/scripts/results/test_xstorycloze_evaluation.json +0 -51
- wisent/examples/scripts/results/test_xstorycloze_pairs.json +0 -14
- wisent/examples/scripts/results/test_xsum_evaluation.json +0 -30
- wisent/examples/scripts/results/test_xsum_pairs.json +0 -8
- wisent/examples/scripts/results/test_xwinograd_evaluation.json +0 -51
- wisent/examples/scripts/results/test_xwinograd_pairs.json +0 -14
- wisent/examples/scripts/results/test_yahoo_answers_topics_evaluation.json +0 -30
- wisent/examples/scripts/results/test_yahoo_answers_topics_pairs.json +0 -8
- {wisent-0.7.379.dist-info → wisent-0.7.701.dist-info}/WHEEL +0 -0
- {wisent-0.7.379.dist-info → wisent-0.7.701.dist-info}/entry_points.txt +0 -0
- {wisent-0.7.379.dist-info → wisent-0.7.701.dist-info}/licenses/LICENSE +0 -0
- {wisent-0.7.379.dist-info → wisent-0.7.701.dist-info}/top_level.txt +0 -0
|
@@ -1,30 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"task_name": "arc_challenge",
|
|
3
|
-
"model_name": "mock",
|
|
4
|
-
"evaluator_name": "log_likelihoods",
|
|
5
|
-
"num_pairs": 1,
|
|
6
|
-
"all_correct": true,
|
|
7
|
-
"pairs": [
|
|
8
|
-
{
|
|
9
|
-
"pair_id": 0,
|
|
10
|
-
"prompt": "Question: Juan and LaKeisha roll a few objects down a ramp. They want to see which object rolls the ...",
|
|
11
|
-
"positive_response": "Record the details of the investigation.",
|
|
12
|
-
"negative_response": "Put the objects in groups.",
|
|
13
|
-
"positive_evaluation": {
|
|
14
|
-
"ground_truth": "TRUTHFUL",
|
|
15
|
-
"confidence": 1.0,
|
|
16
|
-
"expected": "TRUTHFUL",
|
|
17
|
-
"correct": true,
|
|
18
|
-
"details": "Predicted: 'Record the details of the investigation.' (log_prob=-0.500), Expected: 'Record the details of the investigation.'"
|
|
19
|
-
},
|
|
20
|
-
"negative_evaluation": {
|
|
21
|
-
"ground_truth": "UNTRUTHFUL",
|
|
22
|
-
"confidence": 0.0,
|
|
23
|
-
"expected": "UNTRUTHFUL",
|
|
24
|
-
"correct": true,
|
|
25
|
-
"details": "Predicted: 'Record the details of the investigation.' (log_prob=-0.500), Expected: 'Put the objects in groups.'"
|
|
26
|
-
},
|
|
27
|
-
"both_correct": true
|
|
28
|
-
}
|
|
29
|
-
]
|
|
30
|
-
}
|
|
@@ -1,8 +0,0 @@
|
|
|
1
|
-
[
|
|
2
|
-
{
|
|
3
|
-
"pair_id": 0,
|
|
4
|
-
"prompt": "Question: Juan and LaKeisha roll a few objects down a ramp. They want to see which object rolls the farthest. What should they do so they can repeat their investigation?\nA. Put the objects in groups.\nB. Record the details of the investigation.",
|
|
5
|
-
"positive_response": "Record the details of the investigation.",
|
|
6
|
-
"negative_response": "Put the objects in groups."
|
|
7
|
-
}
|
|
8
|
-
]
|
|
@@ -1,30 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"task_name": "arc_easy",
|
|
3
|
-
"model_name": "mock",
|
|
4
|
-
"evaluator_name": "log_likelihoods",
|
|
5
|
-
"num_pairs": 1,
|
|
6
|
-
"all_correct": true,
|
|
7
|
-
"pairs": [
|
|
8
|
-
{
|
|
9
|
-
"pair_id": 0,
|
|
10
|
-
"prompt": "Question: Which technology was developed most recently?\nA. television\nB. cellular telephone...",
|
|
11
|
-
"positive_response": "cellular telephone",
|
|
12
|
-
"negative_response": "television",
|
|
13
|
-
"positive_evaluation": {
|
|
14
|
-
"ground_truth": "TRUTHFUL",
|
|
15
|
-
"confidence": 1.0,
|
|
16
|
-
"expected": "TRUTHFUL",
|
|
17
|
-
"correct": true,
|
|
18
|
-
"details": "Predicted: 'cellular telephone' (log_prob=-0.500), Expected: 'cellular telephone'"
|
|
19
|
-
},
|
|
20
|
-
"negative_evaluation": {
|
|
21
|
-
"ground_truth": "UNTRUTHFUL",
|
|
22
|
-
"confidence": 0.0,
|
|
23
|
-
"expected": "UNTRUTHFUL",
|
|
24
|
-
"correct": true,
|
|
25
|
-
"details": "Predicted: 'cellular telephone' (log_prob=-0.500), Expected: 'television'"
|
|
26
|
-
},
|
|
27
|
-
"both_correct": true
|
|
28
|
-
}
|
|
29
|
-
]
|
|
30
|
-
}
|
|
@@ -1,30 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"task_name": "argument_topic",
|
|
3
|
-
"model_name": "mock",
|
|
4
|
-
"evaluator_name": "log_likelihoods",
|
|
5
|
-
"num_pairs": 1,
|
|
6
|
-
"all_correct": true,
|
|
7
|
-
"pairs": [
|
|
8
|
-
{
|
|
9
|
-
"pair_id": 0,
|
|
10
|
-
"prompt": "Classify the Topic of the following Argument to one of these options: affirmative action, algorithmi...",
|
|
11
|
-
"positive_response": "holocaust denial",
|
|
12
|
-
"negative_response": "affirmative action",
|
|
13
|
-
"positive_evaluation": {
|
|
14
|
-
"ground_truth": "TRUTHFUL",
|
|
15
|
-
"confidence": 1.0,
|
|
16
|
-
"expected": "TRUTHFUL",
|
|
17
|
-
"correct": true,
|
|
18
|
-
"details": "Predicted: 'holocaust denial' (log_prob=-0.500), Expected: 'holocaust denial'"
|
|
19
|
-
},
|
|
20
|
-
"negative_evaluation": {
|
|
21
|
-
"ground_truth": "UNTRUTHFUL",
|
|
22
|
-
"confidence": 0.0,
|
|
23
|
-
"expected": "UNTRUTHFUL",
|
|
24
|
-
"correct": true,
|
|
25
|
-
"details": "Predicted: 'holocaust denial' (log_prob=-0.500), Expected: 'affirmative action'"
|
|
26
|
-
},
|
|
27
|
-
"both_correct": true
|
|
28
|
-
}
|
|
29
|
-
]
|
|
30
|
-
}
|
|
@@ -1,8 +0,0 @@
|
|
|
1
|
-
[
|
|
2
|
-
{
|
|
3
|
-
"pair_id": 0,
|
|
4
|
-
"prompt": "Classify the Topic of the following Argument to one of these options: affirmative action, algorithmic trading, assisted suicide, atheism, austerity regime, blockade of the gaza strip, cancel pride parades, cannabis, capital punishment, collectivism, compulsory voting, cosmetic surgery, cosmetic surgery for minors, embryonic stem cell research, entrapment, executive compensation, factory farming, fast food, fight urbanization, flag burning, foster care, gender-neutral language, guantanamo bay detention camp, holocaust denial, homeopathy, homeschooling, human cloning, intellectual property rights, intelligence tests, journalism, judicial activism, libertarianism, marriage, missionary work, multi-party system, naturopathy, organ trade, payday loans, polygamy, private military companies, prostitution, racial profiling, retirement, safe spaces, school prayer, sex selection, social media, space exploration, stay-at-home dads, student loans, surrogacy, targeted killing, telemarketing, television, the abolition of nuclear weapons, the church of scientology, the development of autonomous cars, the olympic games, the right to keep and bear arms, the three-strikes laws, the use of child actors, the use of economic sanctions, the use of public defenders, the use of school uniform, the vow of celibacy, vocational education, whaling, wikipedia, women in combat, zero-tolerance policy in schools, zoos.\nArgument:\ndenying the holocaust is a valid exercise of freedom of speech\nTopic:",
|
|
5
|
-
"positive_response": "holocaust denial",
|
|
6
|
-
"negative_response": "affirmative action"
|
|
7
|
-
}
|
|
8
|
-
]
|
|
@@ -1,51 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"task_name": "arithmetic",
|
|
3
|
-
"model_name": "mock",
|
|
4
|
-
"evaluator_name": "log_likelihoods",
|
|
5
|
-
"num_pairs": 2,
|
|
6
|
-
"all_correct": true,
|
|
7
|
-
"pairs": [
|
|
8
|
-
{
|
|
9
|
-
"pair_id": 0,
|
|
10
|
-
"prompt": "Question: What is 78 plus 74?\nAnswer:\nA. 153\nB. 152...",
|
|
11
|
-
"positive_response": "152",
|
|
12
|
-
"negative_response": "153",
|
|
13
|
-
"positive_evaluation": {
|
|
14
|
-
"ground_truth": "TRUTHFUL",
|
|
15
|
-
"confidence": 1.0,
|
|
16
|
-
"expected": "TRUTHFUL",
|
|
17
|
-
"correct": true,
|
|
18
|
-
"details": "Predicted: '152' (log_prob=-0.500), Expected: '152'"
|
|
19
|
-
},
|
|
20
|
-
"negative_evaluation": {
|
|
21
|
-
"ground_truth": "UNTRUTHFUL",
|
|
22
|
-
"confidence": 0.0,
|
|
23
|
-
"expected": "UNTRUTHFUL",
|
|
24
|
-
"correct": true,
|
|
25
|
-
"details": "Predicted: '152' (log_prob=-0.500), Expected: '153'"
|
|
26
|
-
},
|
|
27
|
-
"both_correct": true
|
|
28
|
-
},
|
|
29
|
-
{
|
|
30
|
-
"pair_id": 1,
|
|
31
|
-
"prompt": "Question: What is 6204 plus 3166?\nAnswer:\nA. 9371\nB. 9370...",
|
|
32
|
-
"positive_response": "9370",
|
|
33
|
-
"negative_response": "9371",
|
|
34
|
-
"positive_evaluation": {
|
|
35
|
-
"ground_truth": "TRUTHFUL",
|
|
36
|
-
"confidence": 1.0,
|
|
37
|
-
"expected": "TRUTHFUL",
|
|
38
|
-
"correct": true,
|
|
39
|
-
"details": "Predicted: '9370' (log_prob=-0.500), Expected: '9370'"
|
|
40
|
-
},
|
|
41
|
-
"negative_evaluation": {
|
|
42
|
-
"ground_truth": "UNTRUTHFUL",
|
|
43
|
-
"confidence": 0.0,
|
|
44
|
-
"expected": "UNTRUTHFUL",
|
|
45
|
-
"correct": true,
|
|
46
|
-
"details": "Predicted: '9370' (log_prob=-0.500), Expected: '9371'"
|
|
47
|
-
},
|
|
48
|
-
"both_correct": true
|
|
49
|
-
}
|
|
50
|
-
]
|
|
51
|
-
}
|
|
@@ -1,14 +0,0 @@
|
|
|
1
|
-
[
|
|
2
|
-
{
|
|
3
|
-
"pair_id": 0,
|
|
4
|
-
"prompt": "Question: What is 78 plus 74?\nAnswer:\nA. 153\nB. 152",
|
|
5
|
-
"positive_response": "152",
|
|
6
|
-
"negative_response": "153"
|
|
7
|
-
},
|
|
8
|
-
{
|
|
9
|
-
"pair_id": 1,
|
|
10
|
-
"prompt": "Question: What is 6204 plus 3166?\nAnswer:\nA. 9371\nB. 9370",
|
|
11
|
-
"positive_response": "9370",
|
|
12
|
-
"negative_response": "9371"
|
|
13
|
-
}
|
|
14
|
-
]
|
|
@@ -1,30 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"task_name": "asdiv",
|
|
3
|
-
"model_name": "mock",
|
|
4
|
-
"evaluator_name": "log_likelihoods",
|
|
5
|
-
"num_pairs": 1,
|
|
6
|
-
"all_correct": true,
|
|
7
|
-
"pairs": [
|
|
8
|
-
{
|
|
9
|
-
"pair_id": 0,
|
|
10
|
-
"prompt": "Seven red apples and two green apples are in the basket.\nQuestion:How many apples are in the basket?...",
|
|
11
|
-
"positive_response": "9",
|
|
12
|
-
"negative_response": "10",
|
|
13
|
-
"positive_evaluation": {
|
|
14
|
-
"ground_truth": "TRUTHFUL",
|
|
15
|
-
"confidence": 1.0,
|
|
16
|
-
"expected": "TRUTHFUL",
|
|
17
|
-
"correct": true,
|
|
18
|
-
"details": "Predicted: '9' (log_prob=-0.500), Expected: '9'"
|
|
19
|
-
},
|
|
20
|
-
"negative_evaluation": {
|
|
21
|
-
"ground_truth": "UNTRUTHFUL",
|
|
22
|
-
"confidence": 0.0,
|
|
23
|
-
"expected": "UNTRUTHFUL",
|
|
24
|
-
"correct": true,
|
|
25
|
-
"details": "Predicted: '9' (log_prob=-0.500), Expected: '10'"
|
|
26
|
-
},
|
|
27
|
-
"both_correct": true
|
|
28
|
-
}
|
|
29
|
-
]
|
|
30
|
-
}
|
|
@@ -1,30 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"task_name": "assin_entailment",
|
|
3
|
-
"model_name": "mock",
|
|
4
|
-
"evaluator_name": "log_likelihoods",
|
|
5
|
-
"num_pairs": 1,
|
|
6
|
-
"all_correct": true,
|
|
7
|
-
"pairs": [
|
|
8
|
-
{
|
|
9
|
-
"pair_id": 0,
|
|
10
|
-
"prompt": "De acordo com o relat\u00f3rio, foram notificados 6.052 casos suspeitos de dengue, sendo 641 descartados....",
|
|
11
|
-
"positive_response": "De acordo com o relat\u00f3rio, foram notificados 6.052 casos suspeitos de dengue, sendo 641 descartados., certo? Tamb\u00e9m, Do total de casos notificados, 10.768 foram confirmados como dengue e 15.202 descartados.",
|
|
12
|
-
"negative_response": "De acordo com o relat\u00f3rio, foram notificados 6.052 casos suspeitos de dengue, sendo 641 descartados., certo? Sim, Do total de casos notificados, 10.768 foram confirmados como dengue e 15.202 descartados.",
|
|
13
|
-
"positive_evaluation": {
|
|
14
|
-
"ground_truth": "TRUTHFUL",
|
|
15
|
-
"confidence": 1.0,
|
|
16
|
-
"expected": "TRUTHFUL",
|
|
17
|
-
"correct": true,
|
|
18
|
-
"details": "Predicted: 'De acordo com o relat\u00f3rio, foram notificados 6.052 casos suspeitos de dengue, sendo 641 descartados., certo? Tamb\u00e9m, Do total de casos notificados, 10.768 foram confirmados como dengue e 15.202 descartados.' (log_prob=-0.500), Expected: 'De acordo com o relat\u00f3rio, foram notificados 6.052 casos suspeitos de dengue, sendo 641 descartados., certo? Tamb\u00e9m, Do total de casos notificados, 10.768 foram confirmados como dengue e 15.202 descartados.'"
|
|
19
|
-
},
|
|
20
|
-
"negative_evaluation": {
|
|
21
|
-
"ground_truth": "UNTRUTHFUL",
|
|
22
|
-
"confidence": 0.0,
|
|
23
|
-
"expected": "UNTRUTHFUL",
|
|
24
|
-
"correct": true,
|
|
25
|
-
"details": "Predicted: 'De acordo com o relat\u00f3rio, foram notificados 6.052 casos suspeitos de dengue, sendo 641 descartados., certo? Tamb\u00e9m, Do total de casos notificados, 10.768 foram confirmados como dengue e 15.202 descartados.' (log_prob=-0.500), Expected: 'De acordo com o relat\u00f3rio, foram notificados 6.052 casos suspeitos de dengue, sendo 641 descartados., certo? Sim, Do total de casos notificados, 10.768 foram confirmados como dengue e 15.202 descartados.'"
|
|
26
|
-
},
|
|
27
|
-
"both_correct": true
|
|
28
|
-
}
|
|
29
|
-
]
|
|
30
|
-
}
|
|
@@ -1,8 +0,0 @@
|
|
|
1
|
-
[
|
|
2
|
-
{
|
|
3
|
-
"pair_id": 0,
|
|
4
|
-
"prompt": "De acordo com o relat\u00f3rio, foram notificados 6.052 casos suspeitos de dengue, sendo 641 descartados.",
|
|
5
|
-
"positive_response": "De acordo com o relat\u00f3rio, foram notificados 6.052 casos suspeitos de dengue, sendo 641 descartados., certo? Tamb\u00e9m, Do total de casos notificados, 10.768 foram confirmados como dengue e 15.202 descartados.",
|
|
6
|
-
"negative_response": "De acordo com o relat\u00f3rio, foram notificados 6.052 casos suspeitos de dengue, sendo 641 descartados., certo? Sim, Do total de casos notificados, 10.768 foram confirmados como dengue e 15.202 descartados."
|
|
7
|
-
}
|
|
8
|
-
]
|
|
@@ -1,30 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"task_name": "atis",
|
|
3
|
-
"model_name": "mock",
|
|
4
|
-
"evaluator_name": "log_likelihoods",
|
|
5
|
-
"num_pairs": 1,
|
|
6
|
-
"all_correct": true,
|
|
7
|
-
"pairs": [
|
|
8
|
-
{
|
|
9
|
-
"pair_id": 0,
|
|
10
|
-
"prompt": "From the following Text, extract the objects for which the entity type expressed is one of aircraft_...",
|
|
11
|
-
"positive_response": "charlotte: fromloc.city_name, las vegas: toloc.city_name, st. louis: stoploc.city_name",
|
|
12
|
-
"negative_response": "st. louis: stoploc.city_name, las vegas: toloc.city_name, charlotte: fromloc.city_name",
|
|
13
|
-
"positive_evaluation": {
|
|
14
|
-
"ground_truth": "TRUTHFUL",
|
|
15
|
-
"confidence": 1.0,
|
|
16
|
-
"expected": "TRUTHFUL",
|
|
17
|
-
"correct": true,
|
|
18
|
-
"details": "Predicted: 'charlotte: fromloc.city_name, las vegas: toloc.city_name, st. louis: stoploc.city_name' (log_prob=-0.500), Expected: 'charlotte: fromloc.city_name, las vegas: toloc.city_name, st. louis: stoploc.city_name'"
|
|
19
|
-
},
|
|
20
|
-
"negative_evaluation": {
|
|
21
|
-
"ground_truth": "UNTRUTHFUL",
|
|
22
|
-
"confidence": 0.0,
|
|
23
|
-
"expected": "UNTRUTHFUL",
|
|
24
|
-
"correct": true,
|
|
25
|
-
"details": "Predicted: 'charlotte: fromloc.city_name, las vegas: toloc.city_name, st. louis: stoploc.city_name' (log_prob=-0.500), Expected: 'st. louis: stoploc.city_name, las vegas: toloc.city_name, charlotte: fromloc.city_name'"
|
|
26
|
-
},
|
|
27
|
-
"both_correct": true
|
|
28
|
-
}
|
|
29
|
-
]
|
|
30
|
-
}
|
|
@@ -1,8 +0,0 @@
|
|
|
1
|
-
[
|
|
2
|
-
{
|
|
3
|
-
"pair_id": 0,
|
|
4
|
-
"prompt": "From the following Text, extract the objects for which the entity type expressed is one of aircraft_code, airline_code, airline_name, airport_code, airport_name, arrive_date.date_relative, arrive_date.day_name, arrive_date.day_number, arrive_date.month_name, arrive_date.today_relative, arrive_time.end_time, arrive_time.period_mod, arrive_time.period_of_day, arrive_time.start_time, arrive_time.time, arrive_time.time_relative, city_name, class_type, connect, cost_relative, day_name, day_number, days_code, depart_date.date_relative, depart_date.day_name, depart_date.day_number, depart_date.month_name, depart_date.today_relative, depart_date.year, depart_time.end_time, depart_time.period_mod, depart_time.period_of_day, depart_time.start_time, depart_time.time, depart_time.time_relative, economy, fare_amount, fare_basis_code, flight_days, flight_mod, flight_number, flight_stop, flight_time, fromloc.airport_code, fromloc.airport_name, fromloc.city_name, fromloc.state_code, fromloc.state_name, meal, meal_code, meal_description, mod, month_name, or, period_of_day, restriction_code, return_date.date_relative, return_date.day_name, return_date.day_number, return_date.month_name, return_date.today_relative, return_time.period_mod, return_time.period_of_day, round_trip, state_code, state_name, stoploc.airport_name, stoploc.city_name, stoploc.state_code, time, time_relative, today_relative, toloc.airport_code, toloc.airport_name, toloc.city_name, toloc.country_name, toloc.state_code, toloc.state_name, transport_type.\nText:\ni would like to find a flight from charlotte to las vegas that makes a stop in st. louis\nentity type:",
|
|
5
|
-
"positive_response": "charlotte: fromloc.city_name, las vegas: toloc.city_name, st. louis: stoploc.city_name",
|
|
6
|
-
"negative_response": "st. louis: stoploc.city_name, las vegas: toloc.city_name, charlotte: fromloc.city_name"
|
|
7
|
-
}
|
|
8
|
-
]
|
|
@@ -1,30 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"task_name": "babi",
|
|
3
|
-
"model_name": "mock",
|
|
4
|
-
"evaluator_name": "log_likelihoods",
|
|
5
|
-
"num_pairs": 1,
|
|
6
|
-
"all_correct": true,
|
|
7
|
-
"pairs": [
|
|
8
|
-
{
|
|
9
|
-
"pair_id": 0,
|
|
10
|
-
"prompt": "Passage: John travelled to the hallway.\nMary journeyed to the bathroom.\n\nQuestion: Where is John?...",
|
|
11
|
-
"positive_response": "hallway",
|
|
12
|
-
"negative_response": "not hallway",
|
|
13
|
-
"positive_evaluation": {
|
|
14
|
-
"ground_truth": "TRUTHFUL",
|
|
15
|
-
"confidence": 1.0,
|
|
16
|
-
"expected": "TRUTHFUL",
|
|
17
|
-
"correct": true,
|
|
18
|
-
"details": "Predicted: 'hallway' (log_prob=-0.500), Expected: 'hallway'"
|
|
19
|
-
},
|
|
20
|
-
"negative_evaluation": {
|
|
21
|
-
"ground_truth": "UNTRUTHFUL",
|
|
22
|
-
"confidence": 0.0,
|
|
23
|
-
"expected": "UNTRUTHFUL",
|
|
24
|
-
"correct": true,
|
|
25
|
-
"details": "Predicted: 'hallway' (log_prob=-0.500), Expected: 'not hallway'"
|
|
26
|
-
},
|
|
27
|
-
"both_correct": true
|
|
28
|
-
}
|
|
29
|
-
]
|
|
30
|
-
}
|
|
@@ -1,30 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"task_name": "babilong",
|
|
3
|
-
"model_name": "mock",
|
|
4
|
-
"evaluator_name": "log_likelihoods",
|
|
5
|
-
"num_pairs": 1,
|
|
6
|
-
"all_correct": true,
|
|
7
|
-
"pairs": [
|
|
8
|
-
{
|
|
9
|
-
"pair_id": 0,
|
|
10
|
-
"prompt": "Context: \u2013 \u0421\u043c\u043e\u0442\u0440\u0438, \u0414\u0436\u043e, \u2013 \u0441\u043a\u0430\u0437\u0430\u043b\u0430 \u0411\u043e\u043d\u0438\u0442\u0430, \u043f\u043e\u043a\u0430\u0437\u0430\u0432 \u043f\u0430\u043b\u044c\u0446\u0435\u043c. \u041a \u043a\u043e\u043d\u0432\u043e\u044e \u043f\u043e\u0434\u044a\u0435\u0445\u0430\u043b \u0433\u0440\u044f\u0437\u043d\u043e-\u0437\u0435\u043b\u0435\u043d\u044b\u0439 \u043f\u0438\u043a\u0430\u043f \"\u0448...",
|
|
11
|
-
"positive_response": "\u0441\u043f\u0430\u043b\u044c\u043d\u044f",
|
|
12
|
-
"negative_response": "\u0432\u0430\u043d\u043d\u0430\u044f",
|
|
13
|
-
"positive_evaluation": {
|
|
14
|
-
"ground_truth": "TRUTHFUL",
|
|
15
|
-
"confidence": 1.0,
|
|
16
|
-
"expected": "TRUTHFUL",
|
|
17
|
-
"correct": true,
|
|
18
|
-
"details": "Predicted: '\u0441\u043f\u0430\u043b\u044c\u043d\u044f' (log_prob=-0.500), Expected: '\u0441\u043f\u0430\u043b\u044c\u043d\u044f'"
|
|
19
|
-
},
|
|
20
|
-
"negative_evaluation": {
|
|
21
|
-
"ground_truth": "UNTRUTHFUL",
|
|
22
|
-
"confidence": 0.0,
|
|
23
|
-
"expected": "UNTRUTHFUL",
|
|
24
|
-
"correct": true,
|
|
25
|
-
"details": "Predicted: '\u0441\u043f\u0430\u043b\u044c\u043d\u044f' (log_prob=-0.500), Expected: '\u0432\u0430\u043d\u043d\u0430\u044f'"
|
|
26
|
-
},
|
|
27
|
-
"both_correct": true
|
|
28
|
-
}
|
|
29
|
-
]
|
|
30
|
-
}
|
|
@@ -1,8 +0,0 @@
|
|
|
1
|
-
[
|
|
2
|
-
{
|
|
3
|
-
"pair_id": 0,
|
|
4
|
-
"prompt": "Context: \u2013 \u0421\u043c\u043e\u0442\u0440\u0438, \u0414\u0436\u043e, \u2013 \u0441\u043a\u0430\u0437\u0430\u043b\u0430 \u0411\u043e\u043d\u0438\u0442\u0430, \u043f\u043e\u043a\u0430\u0437\u0430\u0432 \u043f\u0430\u043b\u044c\u0446\u0435\u043c. \u041a \u043a\u043e\u043d\u0432\u043e\u044e \u043f\u043e\u0434\u044a\u0435\u0445\u0430\u043b \u0433\u0440\u044f\u0437\u043d\u043e-\u0437\u0435\u043b\u0435\u043d\u044b\u0439 \u043f\u0438\u043a\u0430\u043f \"\u0448\u0435\u0432\u0440\u043e\u043b\u0435\" \u043d\u0430 \u043c\u043e\u0449\u043d\u044b\u0445 \u043a\u043e\u043b\u0435\u0441\u0430\u0445, \u0432\u043e\u0434\u0438\u0442\u0435\u043b\u044c\u0441\u043a\u0430\u044f \u0434\u0432\u0435\u0440\u044c \u0440\u0430\u0441\u043f\u0430\u0445\u043d\u0443\u043b\u0430\u0441\u044c \u0438 \u043e\u0442\u0442\u0443\u0434\u0430 \u0432\u044b\u0431\u0440\u0430\u043b\u0441\u044f \u0414\u0436\u043e \u2013 \u043d\u0430\u0448 \u0434\u0440\u0443\u0433, \u043f\u0430\u0440\u0442\u043d\u0435\u0440 \u043f\u043e \u0431\u0438\u0437\u043d\u0435\u0441\u0443, \u0434\u0430 \u0438 \u043f\u0440\u043e\u0441\u0442\u043e \u043a\u043e\u043b\u043b\u0435\u0433\u0430. \u0420\u043e\u0441\u043b\u044b\u0439, \u043a\u043e\u0440\u043e\u0442\u043a\u043e \u0441\u0442\u0440\u0438\u0436\u0435\u043d\u044b\u0439, \u0432 \u0441\u0432\u0435\u0442\u043b\u044b\u0445 \u0431\u0440\u044e\u043a\u0430\u0445 \u0445\u0430\u043a\u0438 \u0438 \u0447\u0435\u0440\u043d\u043e\u0439 \u0444\u0443\u0442\u0431\u043e\u043b\u043a\u0435. \u041a \u043d\u0435\u043c\u0443 \u043f\u043e\u0434\u043e\u0448\u0435\u043b \u043a\u0442\u043e-\u0442\u043e \u0438\u0437 \u043f\u0440\u0438\u0435\u0445\u0430\u0432\u0448\u0438\u0445 \u0441 \u043a\u043e\u043d\u0432\u043e\u0435\u043c, \u043e\u043d\u0438 \u0437\u0430\u0433\u043e\u0432\u043e\u0440\u0438\u043b\u0438, \u043d\u0435 \u0437\u0430\u0431\u044b\u0432 \u043f\u043e\u0445\u043b\u043e\u043f\u0430\u0442\u044c \u0434\u0440\u0443\u0433 \u0434\u0440\u0443\u0433\u0430 \u043f\u043e \u0441\u043f\u0438\u043d\u0430\u043c. \u2013 \u041e\u043d \u043a\u0430\u043a\u0438\u0435-\u0442\u043e \u0437\u0430\u043f\u0447\u0430\u0441\u0442\u0438 \u0436\u0434\u0430\u043b \u0442\u0430\u043a, \u0447\u0442\u043e \u0441\u043f\u0430\u0442\u044c \u043d\u0435 \u043c\u043e\u0433, \u043a\u0430\u0436\u0435\u0442\u0441\u044f, \u2013 \u043e\u0442\u043e\u0437\u0432\u0430\u043b\u0441\u044f \u044f. \u0421\u043e\u0431\u0435\u0441\u0435\u0434\u043d\u0438\u043a \u0414\u0436\u043e \u043e\u0442\u043a\u0438\u043d\u0443\u043b \u043f\u043e\u043b\u043e\u0433 \u0442\u0435\u043d\u0442\u0430 \u043d\u0430 \u0433\u0440\u0443\u0437\u043e\u0432\u0438\u043a\u0435, \u0437\u0430\u043b\u0435\u0437 \u0432\u043d\u0443\u0442\u0440\u044c \u0438 \u043d\u0430\u0447\u0430\u043b \u043f\u043e\u0434\u0430\u0432\u0430\u0442\u044c \u043e\u0442\u0442\u0443\u0434\u0430 \u043a\u0430\u0440\u0442\u043e\u043d\u043d\u044b\u0435 \u043a\u043e\u0440\u043e\u0431\u043a\u0438, \u0443\u0432\u0435\u0441\u0438\u0441\u0442\u044b\u0435 \u0441 \u0432\u0438\u0434\u0443, \u043a\u043e\u0442\u043e\u0440\u044b\u0435 \u0414\u0436\u043e \u0443\u043a\u043b\u0430\u0434\u044b\u0432\u0430\u043b \u0432 \u043f\u0438\u043a\u0430\u043f. \u2013 \u0422\u0435\u043f\u0435\u0440\u044c \u0443\u0441\u043d\u0435\u0442 \u0441\u043f\u043e\u043a\u043e\u0439\u043d\u043e? \u2013 \u041d\u0430\u0432\u0435\u0440\u043d\u043e\u0435. \u042f \u043e\u0431\u043d\u044f\u043b \u0435\u0435 \u0437\u0430 \u0442\u0430\u043b\u0438\u044e \u0438 \u043f\u0440\u0438\u0442\u044f\u043d\u0443\u043b \u043a \u0441\u0435\u0431\u0435 \u0431\u043b\u0438\u0436\u0435. \u0417\u0430 \u0441\u043f\u0438\u043d\u043e\u0439 \u043e\u0442\u043a\u0440\u044b\u043b\u0430\u0441\u044c \u0434\u0432\u0435\u0440\u044c \u0438 \u0433\u043e\u043b\u043e\u0441 \u043c\u0435\u0434\u0441\u0435\u0441\u0442\u0440\u044b \u0434\u043e\u043a\u0442\u043e\u0440\u0430 \u041c\u0430\u0440\u0442\u0438\u043d\u0435\u0441\u0430, \u0432\u044b\u0441\u043e\u043a\u043e\u0439 \u0438 \u0442\u043e\u0449\u0435\u0439 \u0431\u043b\u043e\u043d\u0434\u0438\u043d\u043a\u0438 \u0425\u044d\u0439\u0437\u0435\u043b, \u043f\u0440\u0438\u0433\u043b\u0430\u0441\u0438\u043b \u0411\u043e\u043d\u0438\u0442\u0443 \u0432\u043e\u0439\u0442\u0438. \u0410\u043b\u0435\u043a\u0441\u0430\u043d\u0434\u0440\u0430 \u043f\u0435\u0440\u0435\u043c\u0435\u0449\u0430\u0435\u0442\u0441\u044f \u0432 \u0441\u043f\u0430\u043b\u044c\u043d\u044e. \u0410 \u044f \u043e\u0441\u0442\u0430\u043b\u0441\u044f \u0436\u0434\u0430\u0442\u044c \u0432 \u043f\u0440\u0438\u0435\u043c\u043d\u043e\u0439, \u0443\u0441\u0435\u0432\u0448\u0438\u0441\u044c \u043d\u0430 \u0434\u0438\u0432\u0430\u043d \u0438 \u0434\u043e\u0441\u0442\u0430\u0432 \u0438\u0437 \u0441\u0443\u043c\u043a\u0438 \u043a\u043d\u0438\u0433\u0443. \u0414\u0443\u043c\u0430\u044e, \u0447\u0442\u043e \u043e\u0436\u0438\u0434\u0430\u043d\u0438\u0435 \u043d\u0435 \u043f\u044f\u0442\u044c \u043c\u0438\u043d\u0443\u0442 \u0437\u0430\u0439\u043c\u0435\u0442. \u041c\u043e\u0436\u043d\u043e \u0431\u044b \u043f\u043e\u0439\u0442\u0438 \u043f\u0440\u043e\u0433\u0443\u043b\u044f\u0442\u044c\u0441\u044f, \u0441 \u0442\u0435\u043c \u0436\u0435 \u0414\u0436\u043e \u043f\u043e\u043e\u0431\u0449\u0430\u0442\u044c\u0441\u044f, \u043d\u043e \u0432\u0434\u0440\u0443\u0433 \u043f\u043e\u043d\u0430\u0434\u043e\u0431\u043b\u044e\u0441\u044c \u0437\u0430\u0447\u0435\u043c-\u043d\u0438\u0431\u0443\u0434\u044c? \u0414\u0432\u0435\u0440\u044c \u0431\u044b\u043b\u0430 \u0442\u043e\u043b\u0441\u0442\u0430\u044f, \u0438\u0437-\u0437\u0430 \u043d\u0435\u0435 \u043d\u0438 \u0437\u0432\u0443\u043a\u0430 \u043d\u0435 \u0434\u043e\u043d\u043e\u0441\u0438\u043b\u043e\u0441\u044c. \u0427\u0438\u0442\u0430\u043b\u043e\u0441\u044c \u0438\u0437-\u0437\u0430 \u0432\u043e\u043b\u043d\u0435\u043d\u0438\u044f \u043f\u043b\u043e\u0445\u043e, \u0432\u0441\u0435 \u0432\u0440\u0435\u043c\u044f \u043b\u043e\u0432\u0438\u043b \u0441\u0435\u0431\u044f \u043d\u0430 \u0442\u043e\u043c, \u0447\u0442\u043e \u043a\u0430\u0436\u0434\u044b\u0439 \u0430\u0431\u0437\u0430\u0446 \u0447\u0438\u0442\u0430\u044e \u043f\u043e \u0434\u0432\u0430-\u0442\u0440\u0438 \u0440\u0430\u0437\u0430, \u0438 \u0432\u0441\u0435 \u0440\u0430\u0432\u043d\u043e \u043d\u0435 \u0437\u0430\u043f\u043e\u043c\u0438\u043d\u0430\u044e, \u043e \u0447\u0435\u043c \u0442\u0430\u043c \u0440\u0435\u0447\u044c. \u0417\u0430\u043a\u0440\u044b\u0432 \u043a\u043d\u0438\u0433\u0443, \u0432\u0441\u0442\u0430\u043b \u0443 \u043e\u043a\u043d\u0430, \u0440\u0430\u0441\u0441\u043c\u0430\u0442\u0440\u0438\u0432\u0430\u044f \u0441\u0443\u0435\u0442\u0443 \u0443 \u043a\u043e\u043d\u0432\u043e\u044f. \u0414\u0436\u043e \u0443\u0436\u0435 \u043d\u0435 \u0431\u044b\u043b\u043e, \u043d\u043e \u0431\u044b\u043b\u043e \u043c\u043d\u043e\u0433\u043e \u0434\u0440\u0443\u0433\u0438\u0445 \u0437\u043d\u0430\u043a\u043e\u043c\u044b\u0445 \u043b\u0438\u0446, \u0438 \u044f \u043d\u0435\u0441\u043a\u043e\u043b\u044c\u043a\u043e \u043c\u0438\u043d\u0443\u0442 \u0438\u0433\u0440\u0430\u043b \u0441\u0430\u043c \u0441 \u0441\u043e\u0431\u043e\u0439 \u0432 \u0438\u0433\u0440\u0443, \u043f\u044b\u0442\u0430\u044f\u0441\u044c \u0443\u0433\u0430\u0434\u0430\u0442\u044c, \u0443 \u043a\u043e\u0433\u043e \u0442\u0430\u043c \u0441\u0435\u0439\u0447\u0430\u0441 \u043a\u0430\u043a\u0438\u0435 \u0434\u0435\u043b\u0430. \u041f\u043e\u0441\u0442\u0435\u043f\u0435\u043d\u043d\u043e \u0442\u043e\u043b\u043f\u0430 \u0440\u0430\u0441\u0441\u0430\u0441\u044b\u0432\u0430\u043b\u0430\u0441\u044c, \u0433\u0440\u0443\u0437\u043e\u0432\u0438\u043a\u0438 \u0440\u0430\u0437\u044a\u0435\u0445\u0430\u043b\u0438\u0441\u044c \u043a\u043e\u043c\u0443 \u043a\u0443\u0434\u0430 \u043d\u0443\u0436\u043d\u043e, \u043e\u0441\u0442\u0430\u043b\u043e\u0441\u044c \u0432\u0441\u0435\u0433\u043e \u043d\u0435\u0441\u043a\u043e\u043b\u044c\u043a\u043e \u0447\u0435\u043b\u043e\u0432\u0435\u043a, \u0432 \u043f\u0435\u0448\u0435\u043c \u043f\u043e\u0440\u044f\u0434\u043a\u0435 \u043d\u0430\u043f\u0440\u0430\u0432\u0438\u0432\u0448\u0438\u0445\u0441\u044f \u0432 \"\u0414\u0436\u0438\u043c\u043c\u0438'\u0437\" \u2013 \u0431\u043b\u0438\u0436\u0430\u0439\u0448\u0438\u0439 \u0431\u0430\u0440. \u041e\u0431\u043c\u044b\u0442\u044c, \u0432\u0438\u0434\u0430\u0442\u044c, \"\u043f\u0440\u0438\u0445\u043e\u0434 \u0432\u0435\u0441\u043d\u044b\". \u0414\u0430, \u0442\u0435\u043f\u0435\u0440\u044c \u0436\u0438\u0437\u043d\u044c \u0441\u043d\u043e\u0432\u0430 \u043e\u0436\u0438\u0432\u0438\u0442\u0441\u044f, \u0430 \u0442\u043e \u043d\u0430 \u0432\u0440\u0435\u043c\u044f \u0441\u0435\u0437\u043e\u043d\u0430 \u0434\u043e\u0436\u0434\u0435\u0439 \u0410\u043b\u0430\u043c\u043e \u043f\u043e\u0433\u0440\u0443\u0437\u0438\u043b\u0441\u044f \u0432 \u0441\u043f\u044f\u0447\u043a\u0443. \u041d\u0435 \u0442\u0430\u043a \u0438 \u043f\u043b\u043e\u0445\u043e \u0431\u044b\u043b\u043e, \u0435\u0441\u043b\u0438 \u0447\u0435\u0441\u0442\u043d\u043e, \u044f \u043f\u0440\u0435\u043a\u0440\u0430\u0441\u043d\u043e \u043e\u0442\u0434\u043e\u0445\u043d\u0443\u043b \u043f\u043e\u0441\u043b\u0435 \u0432\u0441\u0435\u0445 \u0441\u043e\u0431\u044b\u0442\u0438\u0439 \u043f\u0440\u043e\u0448\u043b\u043e\u0433\u043e \u0436\u0430\u0440\u043a\u043e\u0433\u043e \u0441\u0435\u0437\u043e\u043d\u0430, \u043a\u043e\u0442\u043e\u0440\u044b\u0439 \u0434\u043b\u044f \u043c\u0435\u043d\u044f \u043e\u043a\u0430\u0437\u0430\u043b\u0441\u044f \u0434\u0435\u0439\u0441\u0442\u0432\u0438\u0442\u0435\u043b\u044c\u043d\u043e \u0436\u0430\u0440\u043a\u0438\u043c, \u0442\u0430\u043a\u0438\u043c, \u0447\u0442\u043e \u0438 \u0441\u043f\u0435\u0447\u044c\u0441\u044f \u043d\u0435\u0434\u043e\u043b\u0433\u043e. \u041d\u043e \u0432\u0441\u0435 \u0437\u0430\u0432\u0435\u0440\u0448\u0438\u043b\u043e\u0441\u044c \u0431\u043b\u0430\u0433\u043e\u043f\u043e\u043b\u0443\u0447\u043d\u043e, \u0430 \u0432\u044b\u043d\u0443\u0436\u0434\u0435\u043d\u043d\u043e\u0435 \u0434\u043e\u043c\u043e\u0441\u0435\u0434\u0441\u0442\u0432\u043e \u043b\u0438\u0448\u044c \u043f\u0440\u0438\u0432\u0435\u043b\u043e \u043a \u0442\u043e\u043c\u0443, \u0447\u0442\u043e \u0441\u0435\u0439\u0447\u0430\u0441 \u0434\u0443\u0448\u0430 \u0436\u0430\u0436\u0434\u0430\u043b\u0430 \u0434\u0435\u044f\u0442\u0435\u043b\u044c\u043d\u043e\u0441\u0442\u0438. \u0414\u0430 \u0438 \u043f\u043b\u0430\u043d\u043e\u0432 \u043d\u0430 \u044d\u0442\u043e\u0442 \u0441\u0435\u0437\u043e\u043d \u0445\u0432\u0430\u0442\u0430\u043b\u043e, \u0434\u0430\u0436\u0435 \u0432 \u0438\u0445 \u0441\u043f\u0438\u0441\u043e\u043a \u0441\u0442\u0440\u0430\u0448\u043d\u043e \u0437\u0430\u0433\u043b\u044f\u0434\u044b\u0432\u0430\u0442\u044c. \u0414\u043b\u044f \u043d\u0430\u0447\u0430\u043b\u0430 \u043c\u043d\u0435 \u043d\u0430\u0434\u043e \u0432 \u041f\u041f\u0414 \u0441\u0433\u043e\u043d\u044f\u0442\u044c, \u043a\u0430\u043a \u0440\u0430\u0437 \u0441 \u043f\u0435\u0440\u0432\u044b\u043c \u043f\u043e\u043f\u0443\u0442\u043d\u044b\u043c \u043a\u043e\u043d\u0432\u043e\u0435\u043c. \u041f\u043e\u0441\u043a\u043e\u043b\u044c\u043a\u0443 \u044f \"\u0441\u043b\u0443\u0433\u0430 \u0434\u0432\u0443\u0445 \u0433\u043e\u0441\u043f\u043e\u0434\", \u043c\u043d\u0435 \u043d\u0430\u0434\u043e \u0431\u0443\u0434\u0435\u0442 \u0435\u0449\u0435 \u044d\u0442\u0438 \u0441\u0430\u043c\u044b\u0435 \u043f\u043b\u0430\u043d\u044b \u043a\u0430\u043a \u0441 \u043e\u0434\u043d\u0438\u043c \u043d\u0430\u0447\u0430\u043b\u044c\u0441\u0442\u0432\u043e\u043c \u0443\u0442\u0440\u044f\u0441\u0442\u0438, \u0442\u0430\u043a \u0438 \u0441\u043e \u0432\u0442\u043e\u0440\u044b\u043c. \u0410 \u0443\u0447\u0438\u0442\u044b\u0432\u0430\u044f, \u0447\u0442\u043e \u043e\u0442\u043d\u043e\u0448\u0435\u043d\u0438\u044f \u043c\u0435\u0436\u0434\u0443 \u044d\u0442\u0438\u043c\u0438 \"\u0433\u043e\u0441\u043f\u043e\u0434\u0430\u043c\u0438\" \u0441\u043b\u043e\u0436\u043d\u044b\u0435, \u0443\u0442\u0440\u044f\u0441\u0430\u0442\u044c \u0442\u043e\u0436\u0435 \u0431\u0443\u0434\u0435\u0442 \u043d\u0435\u043b\u0435\u0433\u043a\u043e. \u0418 \u0433\u043b\u0430\u0432\u043d\u043e\u0435, \u0442\u043e\u0442 \u0444\u0430\u043a\u0442, \u0447\u0442\u043e \u044f \u0441\u0438\u0436\u0443 \u043c\u0435\u0436\u0434\u0443 \u0434\u0432\u0443\u0445 \u0441\u0442\u0443\u043b\u044c\u0435\u0432, \u043c\u043e\u0435 \u043d\u0430\u0447\u0430\u043b\u044c\u0441\u0442\u0432\u043e \u0441 \u043e\u0431\u0435\u0438\u0445 \u0441\u0442\u043e\u0440\u043e\u043d \u0443\u0434\u043e\u0432\u043b\u0435\u0442\u0432\u043e\u0440\u044f\u0435\u0442. \u041f\u043e\u0445\u043e\u0436\u0435, \u0447\u0442\u043e \u0438\u043c \u043a\u0430\u0436\u0435\u0442\u0441\u044f, \u0447\u0442\u043e \u0442\u0430\u043a\u043e\u0435 \u043f\u043e\u043b\u043e\u0436\u0435\u043d\u0438\u0435 \u0434\u0435\u043b \u043f\u043e\u0437\u0432\u043e\u043b\u0438\u0442 \u0438\u0437\u0431\u0435\u0436\u0430\u0442\u044c \u043a\u043e\u043d\u0444\u043b\u0438\u043a\u0442\u043e\u0432 \u0442\u0430\u043c, \u0433\u0434\u0435 \u0438\u0445 \u043c\u043e\u0436\u043d\u043e \u0438\u0437\u0431\u0435\u0433\u0430\u0442\u044c. \u041d\u0435 \u043b\u0438\u0448\u0435\u043d\u043e \u0441\u043c\u044b\u0441\u043b\u0430, \u043d\u043e \u0438 \u0440\u0438\u0441\u043a\u0438 \u0434\u043b\u044f \u043c\u0435\u043d\u044f \u0438 \u043c\u043e\u0435\u0439 \u043a\u043e\u043c\u0430\u043d\u0434\u044b \u043e\u0442 \u044d\u0442\u043e\u0433\u043e \u043f\u043e\u0432\u044b\u0448\u0430\u044e\u0442\u0441\u044f. \u041f\u043e\u043d\u044f\u0442\u043d\u043e, \u0440\u0430\u0431\u043e\u0442\u0430 \u0443 \u043d\u0430\u0441 \u0442\u0430\u043a\u0430\u044f, \u0447\u0442\u043e \u0431\u0435\u0437 \u0440\u0438\u0441\u043a\u0430 \u043d\u0438\u043a\u0430\u043a, \u043d\u043e \u0445\u043e\u0442\u0435\u043b\u043e\u0441\u044c \u0431\u044b \u0435\u0433\u043e \u043c\u0438\u043d\u0438\u043c\u0438\u0437\u0438\u0440\u043e\u0432\u0430\u0442\u044c \u043f\u043e \u0432\u043e\u0437\u043c\u043e\u0436\u043d\u043e\u0441\u0442\u0438. \u0412 \u043a\u043e\u043d\u0446\u0435-\u043a\u043e\u043d\u0446\u043e\u0432, \u044f \u0436\u0435\u043d\u0430\u0442, \u0438 \u043c\u043d\u0435 \u0435\u0441\u0442\u044c \u0434\u043b\u044f \u043a\u043e\u0433\u043e \u0436\u0438\u0442\u044c. \u0414\u0430, \u0432 \u0441\u0435\u0437\u043e\u043d \u0434\u043e\u0436\u0434\u0435\u0439 \u043c\u044b \u0437\u0430\u0440\u0435\u0433\u0438\u0441\u0442\u0440\u0438\u0440\u043e\u0432\u0430\u043b\u0438 \u0431\u0440\u0430\u043a, \u0442\u043e \u0435\u0441\u0442\u044c \u0432\u0441\u0435 \u0436\u0435 \u043f\u043e\u0436\u0435\u043d\u0438\u043b\u0438\u0441\u044c. \u041f\u043e\u0448\u043b\u0438 \u043a \u043f\u0440\u0435\u043f\u043e\u0434\u043e\u0431\u043d\u043e\u043c\u0443 \u041a\u0443\u0438\u043c\u0431\u0438, \u0438 \"\u0440\u0430\u0441\u043f\u0438\u0441\u0430\u043b\u0438\u0441\u044c\", \u043f\u0440\u0438\u0437\u0432\u0430\u0432 \u0432 \u0441\u0432\u0438\u0434\u0435\u0442\u0435\u043b\u0438 \u0414\u0436\u043e \u0438 \u0435\u0433\u043e \u0434\u043e\u0447\u044c \u0414\u0436\u0435\u0439-\u0414\u0436\u0435\u0439. \u041d\u0435 \u043f\u043e \u0446\u0435\u0440\u043a\u043e\u0432\u043d\u043e\u043c\u0443 \u0440\u0438\u0442\u0443\u0430\u043b\u0443, \u0430 \u043f\u043e \u0433\u0440\u0430\u0436\u0434\u0430\u043d\u0441\u043a\u043e\u043c\u0443, \u043d\u043e \u044d\u0442\u043e \u0442\u043e\u0436\u0435 \u0437\u0434\u0435\u0441\u044c \u0441\u0447\u0438\u0442\u0430\u0435\u0442\u0441\u044f. \u0418 \u043c\u044b \u0437\u0434\u0435\u0441\u044c \u0432\u0441\u0435 \u0436\u0435 \u043d\u0435 \u0441\u043e\u0432\u0441\u0435\u043c \u043c\u0435\u0441\u0442\u043d\u044b\u0435, \u0442\u0430\u043a \u0447\u0442\u043e \u043f\u0440\u0435\u043f\u043e\u0434\u043e\u0431\u043d\u044b\u0439 \u043e\u0442\u043d\u0435\u0441\u0441\u044f \u0441 \u043f\u043e\u043d\u0438\u043c\u0430\u043d\u0438\u0435\u043c. \u0422\u0430\u043a \u0447\u0442\u043e \u043e\u0442\u043d\u043e\u0448\u0435\u043d\u0438\u044f \u043c\u043e\u0438 \u0441 \u0411\u043e\u043d\u0438\u0442\u043e\u0439 \u0442\u0435\u043f\u0435\u0440\u044c \u0430\u0431\u0441\u043e\u043b\u044e\u0442\u043d\u043e \u0437\u0430\u043a\u043e\u043d\u043d\u044b, \u0438 \u043e\u0442 \u043e\u0431\u044f\u0437\u0430\u043d\u043d\u043e\u0441\u0442\u0435\u0439 \u0441\u0443\u043f\u0440\u0443\u0433\u0430, \u043a\u0430\u043a\u0438\u043c\u0438 \u0431\u044b \u043e\u043d\u0438 \u043d\u0438 \u0431\u044b\u043b\u0438, \u043c\u043d\u0435 \u0443\u0436\u0435 \u043d\u0435 \u043e\u0442\u0432\u0435\u0440\u0442\u0435\u0442\u044c\u0441\u044f. \u0414\u0432\u0435\u0440\u044c \u0437\u0430 \u0441\u043f\u0438\u043d\u043e\u0439 \u0440\u0430\u0441\u043f\u0430\u0445\u043d\u0443\u043b\u0430\u0441\u044c \u0438 \u0432 \u043f\u0440\u0438\u0435\u043c\u043d\u043e\u0439 \u043c\u0430\u0442\u0435\u0440\u0438\u0430\u043b\u0438\u0437\u043e\u0432\u0430\u043b\u0430\u0441\u044c \u041c\u0430\u0440\u0438\u044f \u041f\u0438\u043b\u0430\u0440 \u0420\u043e\u0434\u0440\u0438\u0433\u0435\u0441, \u043e\u043d\u0430 \u0436\u0435 \u0411\u043e\u043d\u0438\u0442\u0430, \u0441 \u0432\u0438\u0434\u043e\u043c \u0442\u0430\u0438\u043d\u0441\u0442\u0432\u0435\u043d\u043d\u044b\u043c \u0438 \u044f\u0432\u043d\u043e \u0434\u043e\u0432\u043e\u043b\u044c\u043d\u044b\u043c. \u2013 \u041d\u0443? \u2013 \u0427\u0435\u0442\u044b\u0440\u0435 \u043d\u0435\u0434\u0435\u043b\u0438. \u2013 \u0417\u043e\u043b\u043e\u0442\u0430\u044f! \u2013 \u0440\u0430\u0441\u043a\u0438\u043d\u0443\u043b \u044f \u0440\u0443\u043a\u0438 \u0434\u043b\u044f \u043e\u0431\u044a\u044f\u0442\u0438\u0439. \u2013 \u0414\u0430\u0439 \u043e\u0431\u043d\u0438\u043c\u0443! \u0410 \u043a\u043e\u0433\u0434\u0430 \u043f\u043e\u043b \u0443\u0437\u043d\u0430\u0435\u043c? \u041e\u0442\u0432\u0435\u0442\u0438\u043b \u043f\u043e\u044f\u0432\u0438\u0432\u0448\u0438\u0439\u0441\u044f \u0441\u043b\u0435\u0434\u043e\u043c \u0437\u0430 \u0411\u043e\u043d\u0438\u0442\u043e\u0439 \u0434\u043e\u043a\u0442\u043e\u0440 \u2013 \u043d\u0435\u0432\u044b\u0441\u043e\u043a\u0438\u0439 \u043f\u043e\u043b\u043d\u043e\u0432\u0430\u0442\u044b\u0439 \u043c\u0435\u043a\u0441\u0438\u043a\u0430\u043d\u0435\u0446, \u0432\u044b\u0440\u043e\u0441\u0448\u0438\u0439, \u0432\u043f\u0440\u043e\u0447\u0435\u043c, \u0432 \u0448\u0442\u0430\u0442\u0435 \u041d\u044c\u044e-\u041c\u0435\u043a\u0441\u0438\u043a\u043e \u0438 \u043f\u043e\u043b\u0443\u0447\u0438\u0432\u0448\u0438\u0439 \u0441\u0432\u043e\u0439 \u0434\u0438\u043f\u043b\u043e\u043c \u0432\u0440\u0430\u0447\u0430 \u0432 \u0443\u043d\u0438\u0432\u0435\u0440\u0441\u0438\u0442\u0435\u0442\u0435 \u0410\u0440\u0438\u0437\u043e\u043d\u0430 \u0421\u0442\u044d\u0439\u0442: \u2013 \u041d\u0430\u0434\u043e \u043f\u0440\u0438\u043c\u0435\u0440\u043d\u043e \u0434\u043e \u0434\u0432\u0430\u0434\u0446\u0430\u0442\u044c \u0442\u0440\u0435\u0442\u044c\u0435\u0439 \u043d\u0435\u0434\u0435\u043b\u0438 \u043f\u043e\u0434\u043e\u0436\u0434\u0430\u0442\u044c. \u041f\u043e\u0442\u0435\u0440\u043f\u0438\u0448\u044c? \u2013 \u0414\u043e\u043a, \u0430 \u043a\u0443\u0434\u0430 \u044f \u0434\u0435\u043d\u0443\u0441\u044c? \u2013 \u043e\u0431\u0435\u0440\u043d\u0443\u043b\u0441\u044f \u044f \u043a \u043d\u0435\u043c\u0443, \u043f\u0440\u043e\u0434\u043e\u043b\u0436\u0430\u044f \u043e\u0431\u043d\u0438\u043c\u0430\u0442\u044c \u0443\u0436\u0435 \u043d\u0430\u0447\u0430\u0432\u0448\u0443\u044e \u0432\u044b\u0440\u044b\u0432\u0430\u0442\u044c\u0441\u044f \u0411\u043e\u043d\u0438\u0442\u0443. \u2013 \u0415\u0441\u0442\u044c \u0434\u0440\u0443\u0433\u0438\u0435 \u043e\u043f\u0446\u0438\u0438? \u0422\u043e\u0442 \u043b\u0438\u0448\u044c \u0440\u0430\u0437\u0432\u0435\u043b \u0440\u0443\u043a\u0430\u043c\u0438. \u0410 \u043d\u0430 \u043d\u0435\u0442 \u0438 \u0441\u0443\u0434\u0430 \u043d\u0435\u0442. \u0414\u0430 \u043c\u043d\u0435 \u0432\u0441\u0435 \u0440\u0430\u0432\u043d\u043e, \u0435\u0441\u043b\u0438 \u0447\u0435\u0441\u0442\u043d\u043e, \u043a\u0442\u043e \u0431\u044b \u043d\u0438 \u0431\u044b\u043b. \u0414\u043e \u0441\u043e\u0440\u043e\u043a\u0430 \u043e\u0434\u043d\u043e\u0433\u043e \u0433\u043e\u0434\u0430 \u0434\u043e\u0436\u0438\u043b, \u0430 \u0432\u0441\u0435 \u0432 \u0445\u043e\u043b\u043e\u0441\u0442\u044f\u043a\u0430\u0445 \u0438 \u0431\u0435\u0437 \u0434\u0435\u0442\u0435\u0439. \u041f\u043e \u043c\u043d\u0435 \u0442\u0430\u043a \u0443\u0436\u0435 \u043a\u0440\u0438\u0442\u0438\u0447\u0435\u0441\u043a\u0438\u0439 \u0441\u0440\u043e\u043a, \u043f\u043e\u0437\u0434\u043d\u0435\u0435 \u0438 \u043d\u0435\u043b\u044c\u0437\u044f. \u0412\u043e\u0442 \u0442\u0430\u043a \u043f\u0440\u0438\u043a\u0438\u043d\u0443\u0442\u044c \u0435\u0441\u043b\u0438, \u0447\u0442\u043e \u0432\u043e\u0434\u0438\u0448\u044c \u0442\u044b \u0440\u0435\u0431\u0435\u043d\u043a\u0430 \u0432 \u0448\u043a\u043e\u043b\u0443. \u0423 \u0432\u0441\u0435\u0445 \u043e\u0442\u0446\u044b \u043a\u0430\u043a \u043e\u0442\u0446\u044b, \u0430 \u0443 \u0442\u0432\u043e\u0435\u0433\u043e \u2013 \u0441\u0442\u0430\u0440\u044b\u0439 \u043f\u0435\u0440\u0434\u0443\u043d. \u041e\u043d, \u043f\u043e\u043d\u0438\u043c\u0430\u0435\u0448\u044c, \u0436\u0438\u043b \u0434\u043b\u044f \u0441\u0435\u0431\u044f, \u0430 \u043d\u0430 \u0441\u0442\u0430\u0440\u043e\u0441\u0442\u0438 \u043b\u0435\u0442 \u0441\u043e\u043e\u0431\u0440\u0430\u0437\u0438\u043b, \u0447\u0442\u043e \u0434\u0435\u0442\u0435\u0439 \u0445\u043e\u0447\u0435\u0442. \u2013 \u0422\u044b \u0441\u0447\u0430\u0441\u0442\u043b\u0438\u0432? \u2013 \u0442\u0440\u0435\u0431\u043e\u0432\u0430\u0442\u0435\u043b\u044c\u043d\u043e \u0441\u043f\u0440\u043e\u0441\u0438\u043b\u0430 \u041c\u0430\u0440\u0438\u044f \u041f\u0438\u043b\u0430\u0440 \u0432\u0441\u0435 \u0436\u0435 \u0432\u044b\u0440\u0432\u0430\u0432\u0448\u0438\u0441\u044c. \u2013 \u041d\u0435\u0442, \u0442\u044b \u0432\u0441\u043b\u0443\u0445 \u0441\u043a\u0430\u0436\u0438! \u2013 \u0421\u0447\u0430\u0441\u0442\u043b\u0438\u0432! \u2013 \u0421\u0435\u0437\u043e\u043d \u0434\u043e\u0436\u0434\u0435\u0439 \u0443 \u043d\u0430\u0441 \u0432\u0441\u0435\u0433\u0434\u0430 \u0438\u0437\u043e\u0431\u0438\u043b\u0438\u0435 \u0431\u0435\u0440\u0435\u043c\u0435\u043d\u043d\u043e\u0441\u0442\u0435\u0439 \u0434\u0430\u0435\u0442, \u2013 \u0441\u043a\u0430\u0437\u0430\u043b, \u0443\u043b\u044b\u0431\u043d\u0443\u0432\u0448\u0438\u0441\u044c, \u0434\u043e\u043a\u0442\u043e\u0440. \u0422\u0435\u043f\u0435\u0440\u044c \u0443 \u043c\u0435\u043d\u044f \u043d\u0430\u043f\u043b\u044b\u0432 \u043f\u0430\u0446\u0438\u0435\u043d\u0442\u043e\u0432. \u041d\u0430 \u0443\u043b\u0438\u0446\u0435 \u0441\u043a\u0440\u0438\u043f\u043d\u0443\u043b\u0438 \u0442\u043e\u0440\u043c\u043e\u0437\u0430, \u0443\u043c\u043e\u043b\u043a \u0442\u0430\u0440\u0430\u0445\u0442\u0435\u0432\u0448\u0438\u0439 \u043c\u043e\u0442\u043e\u0440. \u0414\u0432\u0435\u0440\u044c \u0440\u0430\u0441\u043f\u0430\u0445\u043d\u0443\u043b\u0430\u0441\u044c \u0438 \u0432 \u043f\u0440\u0438\u0435\u043c\u043d\u0443\u044e \u0432\u043e\u0448\u043b\u0430 \u0441\u043c\u0443\u0442\u043d\u043e \u0437\u043d\u0430\u043a\u043e\u043c\u0430\u044f \u0436\u0435\u043d\u0449\u0438\u043d\u0430 \u0441 \u0431\u043e\u043b\u044c\u0448\u0438\u043c \u0436\u0438\u0432\u043e\u0442\u043e\u043c. \u042d\u0442\u0430 \u0443\u0436\u0435 \u0442\u043e\u0447\u043d\u043e \u043f\u043e\u043b \u0441\u0432\u043e\u0435\u0433\u043e \u043c\u043b\u0430\u0434\u0435\u043d\u0446\u0430 \u043e\u043f\u0440\u0435\u0434\u0435\u043b\u0438\u043b\u0430, \u0441\u0443\u0434\u044f \u043f\u043e \u0433\u0430\u0431\u0430\u0440\u0438\u0442\u0430\u043c. \u0418 \u043d\u0435 \u0441\u0435\u0437\u043e\u043d \u0434\u043e\u0436\u0434\u0435\u0439 \u043f\u0440\u0438\u0447\u0438\u043d\u043e\u0439 \u0431\u044b\u043b. \u0414\u0430, \u043e\u043d\u0430, \u043a\u0430\u0436\u0435\u0442\u0441\u044f, \u0432 \u0442\u0430\u0443\u043d\u0445\u043e\u043b\u043b\u0435 \u0440\u0430\u0431\u043e\u0442\u0430\u0435\u0442, \u0441\u0435\u043a\u0440\u0435\u0442\u0430\u0440\u0435\u043c, \u044f \u0435\u0435 \u0442\u0430\u043c \u0432\u0438\u0434\u0435\u043b. \u041c\u044b \u043f\u043e\u043f\u0440\u043e\u0449\u0430\u043b\u0438\u0441\u044c \u0441 \u0434\u043e\u043a\u0442\u043e\u0440\u043e\u043c, \u0437\u0430\u0432\u043e\u0434\u0438\u0432\u0448\u0438\u043c \u0441\u043b\u0435\u0434\u0443\u044e\u0449\u0443\u044e \u043f\u0430\u0446\u0438\u0435\u043d\u0442\u043a\u0443 \u0432 \u043a\u0430\u0431\u0438\u043d\u0435\u0442, \u0438 \u0432\u044b\u0448\u043b\u0438 \u043d\u0430 \u0443\u043b\u0438\u0446\u0443. \u2013 \u041d\u0443, \u0432\u043e\u0442 \u0442\u0430\u043a, \u2013 \u0441\u043a\u0430\u0437\u0430\u043b\u0430 \u041c\u0430\u0440\u0438\u044f \u041f\u0438\u043b\u0430\u0440, \u043e\u0431\u0435\u0440\u043d\u0443\u0432\u0448\u0438\u0441\u044c \u043a\u043e \u043c\u043d\u0435. \u2013 \u0422\u0435\u043f\u0435\u0440\u044c \u0432\u0441\u0435 \u0430\u0442\u0440\u0438\u0431\u0443\u0442\u044b \u0441\u0435\u043c\u044c\u0438 \u043d\u0430\u043b\u0438\u0446\u043e. \u041f\u0440\u043e\u0449\u0430\u0439\u0441\u044f \u0441\u043e \u0441\u0432\u043e\u0435\u0439 \u0441\u0432\u043e\u0431\u043e\u0434\u043e\u0439 \u043e\u043a\u043e\u043d\u0447\u0430\u0442\u0435\u043b\u044c\u043d\u043e. \u2013 \u0410\u0433\u0430, \u0441\u0432\u043e\u0431\u043e\u0434\u044b \u0443 \u043c\u0435\u043d\u044f \u0441 \u0442\u043e\u0431\u043e\u0439 \u043f\u0440\u044f\u043c\u043e \u0436\u0443\u0442\u044c \u0441\u043a\u043e\u043b\u044c\u043a\u043e, \u2013 \u0432\u0437\u0434\u043e\u0445\u043d\u0443\u043b \u044f. \u2013 \u041d\u0430 \u043c\u043e\u0439 \u0432\u0437\u0433\u043b\u044f\u0434, \u0442\u0430\u043a \u0441\u043b\u0438\u0448\u043a\u043e\u043c \u043c\u043d\u043e\u0433\u043e. \u041d\u043e \u044d\u0442\u043e \u043c\u044b \u0438\u0441\u043f\u0440\u0430\u0432\u0438\u043c. \u2013 \u041c\u044b? \u2013 \u041a\u043e\u0433\u0434\u0430 \u0440\u0435\u0431\u0435\u043d\u043e\u043a \u0440\u043e\u0434\u0438\u0442\u0441\u044f. \u2013 \u0431\u0443\u0434\u0435\u043c \u0443\u0436\u0435 \"\u043c\u044b\". \u0418 \u0441\u0440\u0430\u0437\u0443 \u0441\u043b\u0435\u0434\u0443\u044e\u0449\u0435\u0433\u043e, \u0447\u0442\u043e\u0431\u044b \u0443\u0436\u0435 \u0440\u0435\u0448\u0438\u0442\u0435\u043b\u044c\u043d\u044b\u0439 \u0447\u0438\u0441\u043b\u0435\u043d\u043d\u044b\u0439 \u043f\u0435\u0440\u0435\u0432\u0435\u0441 \u043d\u0430 \u043d\u0430\u0448\u0435\u0439 \u0441\u0442\u043e\u0440\u043e\u043d\u0435 \u0431\u044b\u043b. \u042f \u043d\u0438\u0447\u0435\u0433\u043e \u043d\u0435 \u043e\u0442\u0432\u0435\u0442\u0438\u043b, \u043b\u0438\u0448\u044c \u0440\u0430\u0437\u0432\u0435\u043b \u0440\u0443\u043a\u0430\u043c\u0438 \u0438 \u0440\u0430\u0441\u043f\u0430\u0445\u043d\u0443\u043b \u043f\u0435\u0440\u0435\u0434 \u043d\u0435\u0439 \u043f\u0430\u0441\u0441\u0430\u0436\u0438\u0440\u0441\u043a\u0443\u044e \u0434\u0432\u0435\u0440\u0446\u0443 \"\u0438\u043b\u0442\u0438\u0441\u0430\" \u2013 \u043d\u0435\u043c\u0435\u0446\u043a\u043e\u0433\u043e \u0432\u043e\u0435\u043d\u043d\u043e\u0433\u043e \u0432\u043d\u0435\u0434\u043e\u0440\u043e\u0436\u043d\u0438\u043a\u0430, \u043c\u0430\u043b\u0435\u043d\u044c\u043a\u043e\u0433\u043e \u0438 \u043d\u0435\u0432\u0437\u0440\u0430\u0447\u043d\u043e\u0433\u043e, \u043a\u043e\u0442\u043e\u0440\u044b\u0439 \u043f\u0440\u0438 \u044d\u0442\u043e\u043c \u043e\u0442\u043b\u0438\u0447\u0430\u043b\u0441\u044f \u0440\u0435\u0434\u043a\u043e\u0439 \u043f\u0440\u043e\u0445\u043e\u0434\u0438\u043c\u043e\u0441\u0442\u044c\u044e \u0438 \u043d\u0435 \u043f\u043e\u0440\u0430\u0436\u0430\u044e\u0449\u0438\u043c \u0432\u043e\u043e\u0431\u0440\u0430\u0436\u0435\u043d\u0438\u0435 \u0430\u043f\u043f\u0435\u0442\u0438\u0442\u043e\u043c \u043a \u0441\u043e\u043b\u044f\u0440\u043a\u0435. \u041a\u0441\u0442\u0430\u0442\u0438, \u043c\u043e\u0436\u043d\u043e \u0443\u0436\u0435. \u041f\u043e\u0436\u0430\u043b\u0443\u0439, \u0438 \u0442\u0435\u043d\u0442 \u0441\u043d\u0438\u043c\u0430\u0442\u044c, \u0434\u043e\u0436\u0434\u0438 \u0437\u0430\u043a\u043e\u043d\u0447\u0438\u043b\u0438\u0441\u044c. \u041a \u0442\u0430\u043a\u043e\u043c\u0443 \u0432\u044b\u0432\u043e\u0434\u0443 \u044f \u043f\u0440\u0438\u0448\u0435\u043b, \u0443\u0441\u0435\u0432\u0448\u0438\u0441\u044c \u0437\u0430 \u0440\u0443\u043b\u044c \u0438 \u043e\u0431\u043d\u0430\u0440\u0443\u0436\u0438\u0432, \u0447\u0442\u043e \u0432\u043e\u0437\u0434\u0443\u0445 \u043f\u043e\u0434 \u0442\u0435\u043d\u0442\u043e\u043c \u0443\u0441\u043f\u0435\u043b \u0440\u0430\u0441\u043a\u0430\u043b\u0438\u0442\u044c\u0441\u044f. \u0425\u043e\u0440\u043e\u0448\u043e \u0431\u044b\u043b\u043e \u0431\u044b \u043e\u043a\u043d\u0430 \u043e\u0442\u043a\u0440\u044b\u0442\u044b\u043c\u0438 \u043e\u0441\u0442\u0430\u0432\u0438\u0442\u044c, \u043d\u043e \u043d\u0435 \u043e\u0442\u043a\u0440\u044b\u0432\u0430\u044e\u0442\u0441\u044f \u0437\u0434\u0435\u0441\u044c \u043e\u043d\u0438, \u043d\u0430\u0434\u043e \u0432\u0441\u044e \u0432\u0435\u0440\u0445\u043d\u044e\u044e \u043f\u043e\u043b\u043e\u0432\u0438\u043d\u043a\u0443 \u0434\u0432\u0435\u0440\u0438 \u0441\u043d\u0438\u043c\u0430\u0442\u044c. \u0427\u0435\u043c \u0434\u043e\u043c\u0430 \u0438 \u0437\u0430\u0439\u043c\u0443\u0441\u044c, \u043f\u043e\u0436\u0430\u043b\u0443\u0439. \u041c\u0430\u0440\u0438\u044f \u0432\u043e\u0437\u0432\u0440\u0430\u0449\u0430\u0435\u0442\u0441\u044f \u043d\u0430 \u043a\u0443\u0445\u043d\u044e. \u041d\u0435\u0431\u043e\u043b\u044c\u0448\u043e\u0439 \u0434\u0438\u0437\u0435\u043b\u0435\u043a \u0440\u044b\u043a\u043d\u0443\u043b \u043f\u043e\u0434 \u043f\u043b\u043e\u0441\u043a\u0438\u043c \u0440\u0435\u0431\u0440\u0438\u0441\u0442\u044b\u043c \u043a\u0430\u043f\u043e\u0442\u043e\u043c \u0432\u0435\u0437\u0434\u0435\u0445\u043e\u0434\u0430, \u043c\u0430\u0448\u0438\u043d\u0430 \u043d\u0435\u0442\u043e\u0440\u043e\u043f\u043b\u0438\u0432\u043e \u0442\u0440\u043e\u043d\u0443\u043b\u0430\u0441\u044c \u0441 \u043c\u0435\u0441\u0442\u0430, \u0445\u0440\u0443\u0441\u0442\u043d\u0443\u043b \u043f\u043e\u0434 \u043f\u043e\u043a\u0440\u044b\u0448\u043a\u0430\u043c\u0438 \u0433\u0440\u0430\u0432\u0438\u0439. \u2013 \u0414\u0430\u0432\u0430\u0439 \u043a \u0423\u0438\u043b\u0441\u043e\u043d\u0443 \u0437\u0430\u0435\u0434\u0435\u043c, \u0445\u043e\u0447\u0443 \u043c\u044f\u0441\u0430 \u043a\u0443\u043f\u0438\u0442\u044c, \u2013 \u0441\u043a\u0430\u0437\u0430\u043b\u0430 \u0411\u043e\u043d\u0438\u0442\u0430. \u2013 \u0421\u0442\u0435\u0439\u043a\u0438 \u043f\u043e\u0436\u0430\u0440\u0438\u043c. \u2013 \u041d\u0435 \u0432\u043e\u043f\u0440\u043e\u0441. \u0427\u0430\u043a \u0423\u0438\u043b\u0441\u043e\u043d \u2013 \u044d\u0442\u043e \u043c\u044f\u0441\u043d\u0438\u043a \u043c\u0435\u0441\u0442\u043d\u044b\u0439. \u0423 \u043d\u0435\u0433\u043e \u0438 \u0441\u0430\u043c\u043e\u0433\u043e \u0444\u0435\u0440\u043c\u0430 \u0435\u0441\u0442\u044c, \u0438 \u0441\u043a\u0443\u043f\u0430\u0435\u0442 \u043e\u043d \u043c\u044f\u0441\u043e \u0443 \u0434\u0440\u0443\u0433\u0438\u0445. \u0412\u044b\u0431\u043e\u0440 \u0443 \u043d\u0435\u0433\u043e \u0445\u043e\u0440\u043e\u0448\u0438\u0439, \u0430 \u043a\u0440\u043e\u043c\u0435 \u0442\u043e\u0433\u043e, \u043c\u043e\u0436\u043d\u043e \u043a\u0443\u043f\u0438\u0442\u044c \u0435\u0449\u0435 \u043c\u043d\u043e\u0433\u043e \u0447\u0435\u0433\u043e \u043f\u043e\u043b\u0435\u0437\u043d\u043e\u0433\u043e. \u0421\u0443\u043f\u0435\u0440\u043c\u0430\u0440\u043a\u0435\u0442\u043e\u0432 \u0437\u0434\u0435\u0441\u044c \u043d\u0435\u0442, \u043d\u0435 \u0434\u043e\u0440\u043e\u0441\u043b\u0438 \u043c\u044b \u0434\u043e \u043d\u0438\u0445, \u0442\u0430\u043a \u0447\u0442\u043e \u0432\u043e\u0442 \u0442\u0430\u043a \u043f\u0440\u043e\u0434\u0443\u043a\u0442\u044b \u0438 \u043f\u043e\u043a\u0443\u043f\u0430\u0435\u043c: \u0447\u0442\u043e \u043f\u0440\u044f\u043c\u043e \u0441 \u0444\u0435\u0440\u043c \u0431\u0435\u0440\u0435\u043c, \u0447\u0442\u043e \u0443 \u043c\u044f\u0441\u043d\u0438\u043a\u0430, \u0447\u0442\u043e \u0443 \u043f\u0435\u043a\u0430\u0440\u044f, \u0447\u0442\u043e \u043c\u043e\u043b\u043e\u0447\u043d\u0438\u043a \u043f\u0440\u0438\u0432\u0435\u0437\u0435\u0442. \u0417\u0430\u0442\u043e \u0441\u0432\u0435\u0436\u0435\u0435 \u0432\u0441\u0435, \u043a\u0430\u043a \u0432\u0441\u043f\u043e\u043c\u043d\u0438\u0448\u044c \u043f\u043e\u0441\u043b\u0435 \u044d\u0442\u043e\u0439 \u0435\u0434\u0443 \u0442\u0443, \u0447\u0442\u043e \"\u0438\u0437 \u043f\u0440\u043e\u0448\u043b\u043e\u0439 \u0436\u0438\u0437\u043d\u0438\", \u0442\u0430\u043a \u0434\u0430\u0436\u0435 \u043f\u043b\u0435\u0432\u0430\u0442\u044c\u0441\u044f \u0445\u043e\u0447\u0435\u0442\u0441\u044f. \u041d\u0430 \u043f\u0435\u0440\u0432\u043e\u043c \u0436\u0435 \u043f\u0435\u0440\u0435\u043a\u0440\u0435\u0441\u0442\u043a\u0435 \u0441\u0432\u0435\u0440\u043d\u0443\u043b \u043d\u0430\u043f\u0440\u0430\u0432\u043e, \u043f\u043e\u043c\u0430\u0445\u0430\u0432 \u0440\u0443\u043a\u043e\u0439 \u0448\u0435\u0434\u0448\u0435\u043c\u0443 \u043d\u0430\u0432\u0441\u0442\u0440\u0435\u0447\u0443 \"\u0421\u044b\u043d\u0443 \u0421\u044d\u043c\u0430\" \u2013 \u043f\u043e\u0434\u0440\u043e\u0441\u0442\u043a\u0443, \u0441\u044b\u043d\u0443 \u043c\u043e\u0435\u0433\u043e \u043f\u0440\u0438\u044f\u0442\u0435\u043b\u044f, \u0430 \u0437\u0430\u043e\u0434\u043d\u043e \u0438 \u043a\u043e\u043d\u043a\u0443\u0440\u0435\u043d\u0442\u0430, \u0432\u043b\u0430\u0434\u0435\u043b\u044c\u0446\u0430 \u0432\u0442\u043e\u0440\u043e\u0433\u043e \u0432 \u044d\u0442\u043e\u043c \u0433\u043e\u0440\u043e\u0434\u0435 \u043e\u0440\u0443\u0436\u0435\u0439\u043d\u043e\u0433\u043e \u043c\u0430\u0433\u0430\u0437\u0438\u043d\u0430. \u0422\u043e\u0442 \u0442\u043e\u0436\u0435 \u043c\u0430\u0445\u043d\u0443\u043b \u0440\u0443\u043a\u043e\u0439 \u0438 \u043f\u0435\u0440\u0435\u0431\u0435\u0436\u0430\u043b \u0434\u043e\u0440\u043e\u0433\u0443 \u0437\u0430 \u043d\u0430\u0448\u0435\u0439 \u043c\u0430\u0448\u0438\u043d\u043e\u0439. \u0414\u043e\u0440\u043e\u0433\u0430 \u0432\u0435\u043b\u0430 \u043d\u0430\u0441 \u043c\u0435\u0436\u0434\u0443 \u0440\u044f\u0434\u0430\u043c\u0438 \u0432\u044b\u0441\u0442\u0440\u043e\u0438\u0432\u0448\u0438\u0445\u0441\u044f \u0432\u0434\u043e\u043b\u044c \u0443\u043b\u0438\u0446\u044b \u0449\u0438\u0442\u043e\u0432\u044b\u0445 \u0434\u043e\u043c\u043e\u0432, \u0432\u044b\u0441\u0442\u0440\u043e\u0435\u043d\u043d\u044b\u0445 \u0432 \u0442\u0438\u043f\u0438\u0447\u043d\u043e \u0430\u043c\u0435\u0440\u0438\u043a\u0430\u043d\u0441\u043a\u043e\u043c \u0441\u0442\u0438\u043b\u0435 \"\u0432\u0435\u0441\u0442\u0435\u0440\u043d\" \u0438 \u043f\u043e \u0442\u0438\u043f\u0438\u0447\u043d\u043e \u0430\u043c\u0435\u0440\u0438\u043a\u0430\u043d\u0441\u043a\u043e\u0439 \u0442\u0435\u0445\u043d\u043e\u043b\u043e\u0433\u0438\u0438, \u0442\u043e \u0435\u0441\u0442\u044c \u0434\u0435\u0448\u0435\u0432\u043e \u0438 \u0431\u044b\u0441\u0442\u0440\u043e, \u0437\u0430\u0442\u043e \u043d\u0435 \u044d\u043a\u043e\u043d\u043e\u043c\u044f \u043c\u0435\u0442\u0440\u0430\u0436. \u041f\u0435\u0440\u0432\u044b\u0435 \u044d\u0442\u0430\u0436\u0438 \u0431\u044b\u043b\u0438 \u0437\u0430\u043d\u044f\u0442\u044b \u043c\u0430\u0433\u0430\u0437\u0438\u043d\u0447\u0438\u043a\u0430\u043c\u0438, \u043a\u0430\u0444\u0435\u0448\u043a\u0430\u043c\u0438, \u043e\u0442\u0434\u0435\u043b\u0435\u043d\u0438\u044f\u043c\u0438 \u0431\u0430\u043d\u043a\u043e\u0432, \u043f\u0430\u0440\u0438\u043a\u043c\u0430\u0445\u0435\u0440\u0441\u043a\u0438\u043c\u0438 \u2013 \u043c\u044b \u043a\u0430\u043a \u0440\u0430\u0437 \u0431\u044b\u043b\u0438 \u0432 \"\u0434\u0435\u043b\u043e\u0432\u043e\u043c \u0446\u0435\u043d\u0442\u0440\u0435\" \u0433\u043e\u0440\u043e\u0434\u043a\u0430 \u0410\u043b\u0430\u043c\u043e. \u041c\u0435\u043b\u044c\u043a\u043d\u0443\u043b\u0438 \u0437\u0434\u0430\u043d\u0438\u044f \u0443\u043f\u0440\u0430\u0432\u043b\u0435\u043d\u0438\u044f \u0448\u0435\u0440\u0438\u0444\u0430 \u0441 \u0432\u044b\u0441\u0442\u0440\u043e\u0438\u0432\u0448\u0438\u043c\u0438\u0441\u044f \u0432\u043e\u0437\u043b\u0435 \u043d\u0435\u0433\u043e \u0447\u0435\u0440\u043d\u043e-\u0431\u0435\u043b\u044b\u043c\u0438 \u0432\u043d\u0435\u0434\u043e\u0440\u043e\u0436\u043d\u0438\u043a\u0430\u043c\u0438, \u0438 \u043c\u0435\u0441\u0442\u043d\u043e\u0433\u043e \u043f\u043e\u0436\u0430\u0440\u043d\u043e\u0433\u043e \u0434\u0435\u043f\u0430\u0440\u0442\u0430\u043c\u0435\u043d\u0442\u0430, \u0441\u0442\u043e\u044f\u043d\u043a\u0430 \u0443 \u043a\u043e\u0442\u043e\u0440\u043e\u0433\u043e \u0431\u044b\u043b\u0430 \u043f\u0443\u0441\u0442\u0430, \u0437\u0430\u0442\u0435\u043c \u043f\u0440\u043e\u0435\u0445\u0430\u043b\u0438 \u0434\u0432\u0443\u0445\u044d\u0442\u0430\u0436\u043d\u043e\u0435 \u0437\u0434\u0430\u043d\u0438\u0435 \u0442\u0430\u0443\u043d-\u0445\u043e\u043b\u043b\u0430 \u0441 \u0440\u0430\u0437\u0432\u0435\u0432\u0430\u044e\u0449\u0438\u043c\u0441\u044f \u043d\u0430\u0434 \u0444\u0430\u0441\u0430\u0434\u043e\u043c \u0444\u043b\u0430\u0433\u043e\u043c \u0441 \u043e\u0434\u0438\u043d\u043e\u043a\u043e\u0439 \u0437\u0432\u0435\u0437\u0434\u043e\u0439. \u0418, \u0432 \u043e\u0431\u0449\u0435\u043c, \u0432\u0435\u0441\u044c \u0446\u0435\u043d\u0442\u0440 \u0433\u043e\u0440\u043e\u0434\u043a\u0430 \u0438 \u043f\u0440\u043e\u0441\u043a\u043e\u0447\u0438\u043b\u0438. \u0410 \u0432\u043e\u043e\u0431\u0449\u0435 \u044f \u0443\u0436\u0435 \u043f\u0440\u0438\u0432\u044b\u043a. \u0418 \u0441\u0432\u043e\u044e \u0431\u044b\u0432\u0448\u0443\u044e \u043c\u043e\u0441\u043a\u043e\u0432\u0441\u043a\u0443\u044e \u0436\u0438\u0437\u043d\u044c \u0432\u0441\u043f\u043e\u043c\u0438\u043d\u0430\u044e \u0441 \u0443\u0436\u0430\u0441\u043e\u043c. \u041e\u0433\u0440\u043e\u043c\u043d\u044b\u0439 \u0433\u043e\u0440\u043e\u0434, \u0448\u0443\u043c, \u0441\u043c\u043e\u0433, \u0433\u0440\u044f\u0437\u044c... \u0434\u0430, \u0442\u0430\u043c \u043a\u0440\u0443\u0433\u043e\u043c \u0430\u0441\u0444\u0430\u043b\u044c\u0442 \u0438 \u043d\u0430 \u043f\u0435\u0440\u0432\u044b\u0439, \u043d\u0435\u0438\u0441\u043a\u0443\u0448\u0435\u043d\u043d\u044b\u0439 \u0432\u0437\u0433\u043b\u044f\u0434, \u0441\u0430\u043c\u0430\u044f \u0433\u0440\u044f\u0437\u044c \u043a\u0430\u043a \u0440\u0430\u0437 \u0437\u0434\u0435\u0441\u044c, \u043f\u043e\u0434 \u043d\u043e\u0433\u0430\u043c\u0438 \u2013 \u043d\u043e \u043f\u043e\u0436\u0438\u0432 \u043d\u0435\u043c\u043d\u043e\u0433\u043e \u0432 \u0410\u043b\u0430\u043c\u043e, \u0438\u043b\u0438 \u0431\u043e\u043b\u044c\u0448\u0438\u043d\u0441\u0442\u0432\u0435 \u0434\u0440\u0443\u0433\u0438\u0445 \u043c\u0430\u043b\u0435\u043d\u044c\u043a\u0438\u0445 \u0433\u043e\u0440\u043e\u0434\u043a\u043e\u0432 \u041d\u043e\u0432\u043e\u0439 \u0417\u0435\u043c\u043b\u0438, \u0442\u044b \u043f\u043e\u043d\u0438\u043c\u0430\u0435\u0448\u044c, \u043a\u0430\u043a \u0442\u044b \u0441 \u044d\u0442\u0438\u043c \u043e\u0448\u0438\u0431\u0430\u043b\u0441\u044f. \u0413\u0440\u044f\u0437\u044c \u2013 \u043e\u043d\u0430 \u0432\u0441\u0435 \u0436\u0435 \u0442\u0430\u043c. \u0410 \u0437\u0434\u0435\u0441\u044c... \u0437\u0434\u0435\u0441\u044c \u043f\u0440\u043e\u0441\u0442\u043e \u043d\u0435\u0442 \u043f\u043e\u043a\u0430 \u0432\u043e\u0437\u043c\u043e\u0436\u043d\u043e\u0441\u0442\u0438 \u0432\u0441\u0435 \u0437\u0430\u043c\u043e\u0441\u0442\u0438\u0442\u044c. \u041d\u043e \u043a\u043e\u0433\u0434\u0430-\u043d\u0438\u0431\u0443\u0434\u044c \u0431\u0443\u0434\u0435\u0442. \u0410 \u043d\u0435 \u0431\u0443\u0434\u0435\u0442 \u2013 \u0442\u0430\u043a \u0438 \u0447\u0435\u0440\u0442 \u0441 \u043d\u0435\u0439, \u043c\u044b \u0432 \u0434\u043e\u0436\u0434\u0438 \u0438 \u0431\u043e\u0442\u0438\u043d\u043a\u0438 \u043d\u0430\u0434\u0435\u043d\u0435\u043c \u043f\u043e\u0432\u044b\u0448\u0435, \u0438 \u0432\u044b\u0442\u0440\u0435\u043c \u0438\u0445 \u0445\u043e\u0440\u043e\u0448\u043e \u043f\u0435\u0440\u0435\u0434 \u0442\u0435\u043c, \u043a\u0430\u043a \u043a\u0443\u0434\u0430-\u0442\u043e \u0432\u043e\u0439\u0442\u0438. \u041c\u0430\u0433\u0430\u0437\u0438\u043d\u0447\u0438\u043a \u0423\u0438\u043b\u0441\u043e\u043d\u0430 \u0437\u0430\u043d\u0438\u043c\u0430\u043b \u043f\u043e\u043b\u043e\u0432\u0438\u043d\u0443 \u043f\u0435\u0440\u0432\u043e\u0433\u043e \u044d\u0442\u0430\u0436\u0430 \u0443\u0433\u043b\u043e\u0432\u043e\u0433\u043e \u0434\u043e\u043c\u0430, \u0447\u0442\u043e \u0440\u0430\u0441\u043f\u043e\u043b\u043e\u0436\u0438\u043b\u0441\u044f \u043d\u0430 \u043f\u0435\u0440\u0435\u0441\u0435\u0447\u0435\u043d\u0438\u0438 \u0443\u043b\u0438\u0446\u044b \u041f\u0438\u043e\u043d\u0435\u0440\u043e\u0432 (\u043d\u0435 \u043f\u0443\u0442\u0430\u0442\u044c \u0441 \u044e\u043d\u044b\u043c\u0438 \u043b\u0435\u043d\u0438\u043d\u0446\u0430\u043c\u0438) \u0438 \u0423\u0438\u043b\u044c\u044f\u043c\u0430 \u041f\u0435\u0440\u0440\u0438, \u043d\u0430\u0437\u0432\u0430\u043d\u043d\u043e\u0439 \u0432 \u0447\u0435\u0441\u0442\u044c \u043c\u0435\u0441\u0442\u043d\u043e\u0433\u043e \u0436\u0438\u0442\u0435\u043b\u044f, \u043e\u0442\u043b\u0438\u0447\u0438\u0432\u0448\u0435\u0433\u043e\u0441\u044f \u043a\u043e\u0433\u0434\u0430-\u0442\u043e \u0432 \u043e\u0431\u043e\u0440\u043e\u043d\u0435 \u0433\u043e\u0440\u043e\u0434\u043a\u0430 \u0432\u043e \u0432\u0440\u0435\u043c\u044f \u043d\u0430\u0431\u0435\u0433\u0430 \u0431\u043e\u043b\u044c\u0448\u043e\u0439 \u0431\u0430\u043d\u0434\u044b. \u0423 \u0432\u0445\u043e\u0434\u0430 \u0432 \u043c\u0430\u0433\u0430\u0437\u0438\u043d \u0441\u0442\u043e\u044f\u043b\u0438 \u0434\u0432\u0430 \u0432\u0435\u043b\u043e\u0441\u0438\u043f\u0435\u0434\u0430 \u0441 \u043a\u043e\u0440\u0437\u0438\u043d\u043a\u0430\u043c\u0438, \u0438 \u0441\u0442\u0430\u0440\u0435\u043d\u044c\u043a\u0438\u0439 \u0444\u043e\u0440\u0434\u043e\u0432\u0441\u043a\u0438\u0439 \u043f\u0438\u043a\u0430\u043f\u0447\u0438\u043a, \u0432\u043f\u0440\u043e\u0447\u0435\u043c, \u043f\u0435\u0440\u0435\u0434\u0435\u043b\u0430\u043d\u043d\u044b\u0439\n\nQuestion: \u0413\u0434\u0435 \u0410\u043b\u0435\u043a\u0441\u0430\u043d\u0434\u0440\u0430?",
|
|
5
|
-
"positive_response": "\u0441\u043f\u0430\u043b\u044c\u043d\u044f",
|
|
6
|
-
"negative_response": "\u0432\u0430\u043d\u043d\u0430\u044f"
|
|
7
|
-
}
|
|
8
|
-
]
|
|
@@ -1,30 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"task_name": "bangla_mmlu",
|
|
3
|
-
"model_name": "mock",
|
|
4
|
-
"evaluator_name": "log_likelihoods",
|
|
5
|
-
"num_pairs": 1,
|
|
6
|
-
"all_correct": true,
|
|
7
|
-
"pairs": [
|
|
8
|
-
{
|
|
9
|
-
"pair_id": 0,
|
|
10
|
-
"prompt": "Question: \u09b8\u09cd\u09a8\u09be\u0995\u09c7\u09a8\u09ac\u09be\u09b0\u09cd\u0997 \u098f\u09ac\u0982 \u099f\u09ae\u09b2\u09bf\u09a8\u09b8\u09a8 (2014)-\u098f\u09b0 \u09ae\u09a4\u09be\u09a8\u09c1\u09b8\u09be\u09b0\u09c7 \u09b8\u0982\u09b8\u09cd\u09a5\u09be\u0997\u09a4 \u09b8\u09cd\u09ac\u099a\u09cd\u099b\u09a4\u09be\u09b0 \u0997\u09c1\u09a3\u09ae\u09be\u09a8 \u09ae\u09c2\u09b2\u09cd\u09af\u09be\u09af\u09bc\u09a8\u09c7\u09b0 \u099c\u09a8\u09cd\u09af \u0995\u09cb...",
|
|
11
|
-
"positive_response": "\u09aa\u09cd\u09b0\u0995\u09be\u09b6, \u09b8\u09cd\u09aa\u09b7\u09cd\u099f\u09a4\u09be, \u09a8\u09bf\u09b0\u09cd\u09ad\u09c1\u09b2\u09a4\u09be",
|
|
12
|
-
"negative_response": "\u09aa\u09cd\u09b0\u09b8\u09cd\u09a5, \u09b8\u09cd\u09aa\u09b7\u09cd\u099f\u09a4\u09be, \u09a8\u09bf\u09b0\u09cd\u09ad\u09c1\u09b2\u09a4\u09be",
|
|
13
|
-
"positive_evaluation": {
|
|
14
|
-
"ground_truth": "TRUTHFUL",
|
|
15
|
-
"confidence": 1.0,
|
|
16
|
-
"expected": "TRUTHFUL",
|
|
17
|
-
"correct": true,
|
|
18
|
-
"details": "Predicted: '\u09aa\u09cd\u09b0\u0995\u09be\u09b6, \u09b8\u09cd\u09aa\u09b7\u09cd\u099f\u09a4\u09be, \u09a8\u09bf\u09b0\u09cd\u09ad\u09c1\u09b2\u09a4\u09be' (log_prob=-0.500), Expected: '\u09aa\u09cd\u09b0\u0995\u09be\u09b6, \u09b8\u09cd\u09aa\u09b7\u09cd\u099f\u09a4\u09be, \u09a8\u09bf\u09b0\u09cd\u09ad\u09c1\u09b2\u09a4\u09be'"
|
|
19
|
-
},
|
|
20
|
-
"negative_evaluation": {
|
|
21
|
-
"ground_truth": "UNTRUTHFUL",
|
|
22
|
-
"confidence": 0.0,
|
|
23
|
-
"expected": "UNTRUTHFUL",
|
|
24
|
-
"correct": true,
|
|
25
|
-
"details": "Predicted: '\u09aa\u09cd\u09b0\u0995\u09be\u09b6, \u09b8\u09cd\u09aa\u09b7\u09cd\u099f\u09a4\u09be, \u09a8\u09bf\u09b0\u09cd\u09ad\u09c1\u09b2\u09a4\u09be' (log_prob=-0.500), Expected: '\u09aa\u09cd\u09b0\u09b8\u09cd\u09a5, \u09b8\u09cd\u09aa\u09b7\u09cd\u099f\u09a4\u09be, \u09a8\u09bf\u09b0\u09cd\u09ad\u09c1\u09b2\u09a4\u09be'"
|
|
26
|
-
},
|
|
27
|
-
"both_correct": true
|
|
28
|
-
}
|
|
29
|
-
]
|
|
30
|
-
}
|
|
@@ -1,8 +0,0 @@
|
|
|
1
|
-
[
|
|
2
|
-
{
|
|
3
|
-
"pair_id": 0,
|
|
4
|
-
"prompt": "Question: \u09b8\u09cd\u09a8\u09be\u0995\u09c7\u09a8\u09ac\u09be\u09b0\u09cd\u0997 \u098f\u09ac\u0982 \u099f\u09ae\u09b2\u09bf\u09a8\u09b8\u09a8 (2014)-\u098f\u09b0 \u09ae\u09a4\u09be\u09a8\u09c1\u09b8\u09be\u09b0\u09c7 \u09b8\u0982\u09b8\u09cd\u09a5\u09be\u0997\u09a4 \u09b8\u09cd\u09ac\u099a\u09cd\u099b\u09a4\u09be\u09b0 \u0997\u09c1\u09a3\u09ae\u09be\u09a8 \u09ae\u09c2\u09b2\u09cd\u09af\u09be\u09af\u09bc\u09a8\u09c7\u09b0 \u099c\u09a8\u09cd\u09af \u0995\u09cb\u09a8 \u09a4\u09bf\u09a8\u099f\u09bf \u09ae\u09be\u09a8\u09a6\u09a3\u09cd\u09a1 \u09ac\u09cd\u09af\u09ac\u09b9\u09be\u09b0 \u0995\u09b0\u09be \u09af\u09c7\u09a4\u09c7 \u09aa\u09be\u09b0\u09c7?\nA. \u09aa\u09cd\u09b0\u09b8\u09cd\u09a5, \u09b8\u09cd\u09aa\u09b7\u09cd\u099f\u09a4\u09be, \u09a8\u09bf\u09b0\u09cd\u09ad\u09c1\u09b2\u09a4\u09be\nB. \u09aa\u09cd\u09b0\u0995\u09be\u09b6, \u09b8\u09cd\u09aa\u09b7\u09cd\u099f\u09a4\u09be, \u09a8\u09bf\u09b0\u09cd\u09ad\u09c1\u09b2\u09a4\u09be",
|
|
5
|
-
"positive_response": "\u09aa\u09cd\u09b0\u0995\u09be\u09b6, \u09b8\u09cd\u09aa\u09b7\u09cd\u099f\u09a4\u09be, \u09a8\u09bf\u09b0\u09cd\u09ad\u09c1\u09b2\u09a4\u09be",
|
|
6
|
-
"negative_response": "\u09aa\u09cd\u09b0\u09b8\u09cd\u09a5, \u09b8\u09cd\u09aa\u09b7\u09cd\u099f\u09a4\u09be, \u09a8\u09bf\u09b0\u09cd\u09ad\u09c1\u09b2\u09a4\u09be"
|
|
7
|
-
}
|
|
8
|
-
]
|
|
@@ -1,30 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"task_name": "banking77",
|
|
3
|
-
"model_name": "mock",
|
|
4
|
-
"evaluator_name": "log_likelihoods",
|
|
5
|
-
"num_pairs": 1,
|
|
6
|
-
"all_correct": true,
|
|
7
|
-
"pairs": [
|
|
8
|
-
{
|
|
9
|
-
"pair_id": 0,
|
|
10
|
-
"prompt": "Classify the Intent of the following Utterance to one of these options: activate my card, age limit,...",
|
|
11
|
-
"positive_response": "top up by cash or cheque",
|
|
12
|
-
"negative_response": "activate my card",
|
|
13
|
-
"positive_evaluation": {
|
|
14
|
-
"ground_truth": "TRUTHFUL",
|
|
15
|
-
"confidence": 1.0,
|
|
16
|
-
"expected": "TRUTHFUL",
|
|
17
|
-
"correct": true,
|
|
18
|
-
"details": "Predicted: 'top up by cash or cheque' (log_prob=-0.500), Expected: 'top up by cash or cheque'"
|
|
19
|
-
},
|
|
20
|
-
"negative_evaluation": {
|
|
21
|
-
"ground_truth": "UNTRUTHFUL",
|
|
22
|
-
"confidence": 0.0,
|
|
23
|
-
"expected": "UNTRUTHFUL",
|
|
24
|
-
"correct": true,
|
|
25
|
-
"details": "Predicted: 'top up by cash or cheque' (log_prob=-0.500), Expected: 'activate my card'"
|
|
26
|
-
},
|
|
27
|
-
"both_correct": true
|
|
28
|
-
}
|
|
29
|
-
]
|
|
30
|
-
}
|
|
@@ -1,8 +0,0 @@
|
|
|
1
|
-
[
|
|
2
|
-
{
|
|
3
|
-
"pair_id": 0,
|
|
4
|
-
"prompt": "Classify the Intent of the following Utterance to one of these options: activate my card, age limit, apple pay or google pay, atm support, automatic top up, balance not updated after bank transfer, balance not updated after cheque or cash deposit, beneficiary not allowed, cancel transfer, card about to expire, card acceptance, card arrival, card delivery estimate, card linking, card not working, card payment fee charged, card payment not recognised, card payment wrong exchange rate, card swallowed, cash withdrawal charge, cash withdrawal not recognised, change pin, compromised card, contactless not working, country support, declined card payment, declined cash withdrawal, declined transfer, direct debit payment not recognised, disposable card limits, edit personal details, exchange charge, exchange rate, exchange via app, extra charge on statement, failed transfer, fiat currency support, get disposable virtual card, get physical card, getting spare card, getting virtual card, lost or stolen card, lost or stolen phone, order physical card, passcode forgotten, pending card payment, pending cash withdrawal, pending top up, pending transfer, pin blocked, receiving money, Refund not showing up, request refund, reverted card payment?, supported cards and currencies, terminate account, top up by bank transfer charge, top up by card charge, top up by cash or cheque, top up failed, top up limits, top up reverted, topping up by card, transaction charged twice, transfer fee charged, transfer into account, transfer not received by recipient, transfer timing, unable to verify identity, verify my identity, verify source of funds, verify top up, virtual card not working, visa or mastercard, why verify identity, wrong amount of cash received, wrong exchange rate for cash withdrawal.\nUtterance:\ncan i deposit a check here?\nIntent:",
|
|
5
|
-
"positive_response": "top up by cash or cheque",
|
|
6
|
-
"negative_response": "activate my card"
|
|
7
|
-
}
|
|
8
|
-
]
|
|
@@ -1,14 +0,0 @@
|
|
|
1
|
-
[
|
|
2
|
-
{
|
|
3
|
-
"pair_id": 0,
|
|
4
|
-
"prompt": "Question: Hegazkinak kontrola galdu du. Ai, ene, .....\nA. gurea egin digu!\nB. gureak egin du!",
|
|
5
|
-
"positive_response": "gureak egin du!",
|
|
6
|
-
"negative_response": "gurea egin digu!"
|
|
7
|
-
},
|
|
8
|
-
{
|
|
9
|
-
"pair_id": 1,
|
|
10
|
-
"prompt": "Question: Larrialdietako Harrerak ingresuen harreraren bulego delegatu gisafuntzionatzen du:\nA. Ezingo du jardun bulego delegatu gisa, jaiegunetan izan ezik.\nB. Egunez.",
|
|
11
|
-
"positive_response": "Egunez.",
|
|
12
|
-
"negative_response": "Ezingo du jardun bulego delegatu gisa, jaiegunetan izan ezik."
|
|
13
|
-
}
|
|
14
|
-
]
|